AI-Model-Zoo

Vitis AI Model Zoo

Introduction

This repository includes optimized deep learning models to speed up the deployment of deep learning inference on Xilinx™ platforms. These models cover different applications, including but not limited to ADAS/AD, medical, video surveillance, robotics, data center, etc. You can get started with these free pre-trained models to enjoy the benefits of deep learning acceleration.

Vitis AI 1.4 Model Zoo New Features！

1.New 16 models and total 108 models with diverse deep learning frameworks (Caffe, TensorFlow, TensorFlow 2 and PyTorch) in Vitis AI 1.4.

2.The diversity of AI Model Zoo is significantly improved compared to Vitis AI 1.3.

a.For autonomous driving and ADAS, provide 4D radar detection, image-lidar sensor fusion, surround-view 3D detection, upgraded 3D segmentation and upgraded multi-task models.
b.For medical and industrial vision, provide depth estimation, RGB-D segmentation, super resolution and many other reference models.

3.EoU enhancement: provide automated download script, freely select the version according to model name and hardware platform.

Model Details

The following table includes comprehensive information about all models, including application, framework, training and validation dataset, input size, computation as well as float and quantized precision.

Click here to view details

No.	Application	Name	Framework	Input Size	OPS per image	Training Set	Val Set	Float Accuracy (Top1/ Top5)	Quantized Accuracy (Top1/Top5)
1	Image Classification	cf_resnet50_imagenet_224_224_7.7G_1.4	caffe	224*224	7.7G	ImageNet Train	ImageNet Val	0.7444/0.9185	0.7335/0.9130
2	Image Classifiction	cf_resnet18_imagenet_224_224_3.65G_1.4	caffe	224*224	3.65G	ImageNet Train	ImageNet Val	0.6844/0.8864	0.6699/0.8826
3	Image Classification	cf_inceptionv1_imagenet_224_224_3.16G_1.4	caffe	224*224	3.16G	ImageNet Train	ImageNet Val	0.7030/0.8971	0.6964/0.8942
4	Image Classification	cf_inceptionv2_imagenet_224_224_4G_1.4	caffe	224*224	4G	ImageNet Train	ImageNet Val	0.7275/0.9111	0.7166/0.9029
5	Image Classification	cf_inceptionv3_imagenet_299_299_11.4G_1.4	caffe	299*299	11.4G	ImageNet Train	ImageNet Val	0.7701/0.9329	0.7622/0.9302
6	Image Classification	cf_inceptionv4_imagenet_299_299_24.5G_1.4	caffe	299*299	24.5G	ImageNet Train	ImageNet Val	0.7958/0.9470	0.7899/0.9444
7	Image Classification	cf_mobilenetv2_imagenet_224_224_0.59G_1.4	caffe	224*224	608M	ImageNet Train	ImageNet Val	0.6527/0.8643	0.6399/0.8540
8	Image Classifiction	cf_squeezenet_imagenet_227_227_0.76G_1.4	caffe	227*227	0.76G	ImageNet Train	ImageNet Val	0.5432/0.7824	0.5270/0.7816
9	ADAS Pedstrain Detection	cf_ssdpedestrian_coco_360_640_0.97_5.9G_1.4	caffe	360*640	5.9G	coco2014_train_person and crowndhuman	coco2014_val_person	0.5903	0.5857
10	Object Detection	cf_refinedet_coco_360_480_123G_1.4	caffe	360*480	123G	coco2014_train_person	coco2014_val_person	0.6928	0.7042
11	Object Detection	cf_refinedet_coco_360_480_0.8_25G_1.4	caffe	360*480	25G	coco2014_train_person	coco2014_val_person	0.6794	0.6790
12	Object Detection	cf_refinedet_coco_360_480_0.92_10.10G_1.4	caffe	360*480	10.10G	coco2014_train_person	coco2014_val_person	0.6489	0.6487
13	Object Detection	cf_refinedet_coco_360_480_0.96_5.08G_1.4	caffe	360*480	5.08G	coco2014_train_person	coco2014_val_person	0.6120	0.6112
14	ADAS Vehicle Detection	cf_ssdadas_bdd_360_480_0.95_6.3G_1.4	caffe	360*480	6.3G	bdd100k + private data	bdd100k + private data	0.4207	0.4200
15	Traffic Detection	cf_ssdtraffic_360_480_0.9_11.6G_1.4	caffe	360*480	11.6G	private data	private data	0.5982	0.5921
16	ADAS Lane Detection	cf_VPGnet_caltechlane_480_640_0.99_2.5G_1.4	caffe	480*640	2.5G	caltech lanes	caltech lanes	0.8864 (F1-score)	0.8882 (F1-score)
17	Object Detection	cf_ssdmobilenetv2_bdd_360_480_6.57G_1.4	caffe	360*480	6.57G	bdd100k train	bdd100k val	0.3052	0.2752
18	ADAS Segmentation	cf_fpn_cityscapes_256_512_8.9G_1.4	caffe	256*512	8.9G	Cityscapes gtFineTrain	Cityscapes Val	0.5669	0.5607
19	Pose Estimation	cf_SPnet_aichallenger_224_128_0.54G_1.4	caffe	224*128	548.6M	ai_challenger	ai_challenger	0.9000 (PCKh0.5)	0.8964 (PCKh0.5)
20	Pose Estimation	cf_openpose_aichallenger_368_368_0.3_189.7G_1.4	caffe	368*368	189.7G	ai_challenger	ai_challenger	0.4507 (OKs)	0.4429 (Oks)
21	Face Detection	cf_densebox_wider_320_320_0.49G_1.4	caffe	320*320	0.49G	wider_face	FDDB	0.8833	0.8783
22	Face Detection	cf_densebox_wider_360_640_1.11G_1.4	caffe	360*640	1.11G	wider_face	FDDB	0.8931	0.8922
23	Face Recognition	cf_landmark_celeba_96_72_0.14G_1.4	caffe	96*72	0.14G	celebA	processed helen	0.1952 (L2 loss)	0.1971 (L2 loss)
24	Re-identification	cf_reid_market1501_160_80_0.95G_1.4	caffe	160*80	0.95G	Market1501+CUHK03	Market1501	mAP:0.5660 Rank-1:0.7800	mAP:0.5590 Rank-1:0.7760
25	Detection+Segmentation	cf_multitask_bdd_288_512_14.8G_1.4	caffe	288*512	14.8G	BDD100K+Cityscapes	BDD100K+Cityscapes	mAP:0.2228 mIOU:0.4088	mAP:0.2140 mIOU:0.4058
26	Object Detection	dk_yolov3_bdd_288_512_53.7G_1.4	darknet	288*512	53.7G	bdd100k	bdd100k	0.5058	0.4910
27	Object Detection	dk_yolov3_cityscapes_256_512_0.9_5.46G_1.4	darknet	256*512	5.46G	Cityscapes	Cityscapes	0.5520	0.5300
28	Object Detection	dk_yolov3_voc_416_416_65.42G_1.4	darknet	416*416	65.42G	voc07+12_trainval	voc07_test	0.8240	0.8130
29	Object Detection	dk_yolov2_voc_448_448_34G_1.4	darknet	448*448	34G	voc07+12_trainval	voc07_test	0.7845	0.7740
30	Object Detection	dk_yolov2_voc_448_448_0.66_11.56G_1.4	darknet	448*448	11.56G	voc07+12_trainval	voc07_test	0.7700	0.7610
31	Object Detection	dk_yolov2_voc_448_448_0.71_9.86G_1.4	darknet	448*448	9.86G	voc07+12_trainval	voc07_test	0.7670	0.7540
32	Object Detection	dk_yolov2_voc_448_448_0.77_7.82G_1.4	darknet	448*448	7.82G	voc07+12_trainval	voc07_test	0.7576	0.7482
33	Face Recognition	cf_facerec-resnet20_112_96_3.5G_1.4	caffe	112*96	3.5G	private data	private data	0.9610 (1e-06)	0.9480 (1e-06)
34	Face Recognition	cf_facerec-resnet64_112_96_11G_1.4	caffe	112*96	11G	private data	private data	0.9830 (1e-06)	0.9820 (1e-06)
35	Medical Segmentation	cf_FPN-resnet18_EDD_320_320_45.3G_1.4	caffe	320*320	45.30G	EDD_seg	EDD_seg	mean dice=0.8203 mean jaccard=0.7925 F2-score=0.8075	mean dice=0.8055 mean jaccard=0.7772 F2- score=0.7925
36	Plate Detection	cf_plate-detection_320_320_0.49G_1.4	caffe	320*320	0.49G	private data	private data	Recall1: 0.9660	Recall1: 0.9650
37	Plate Recognition	cf_plate-recognition_96_288_1.75G_1.4	caffe	96*288	1.75G	private data	private data	plate number:99.51% plate color:100%	plate number:99.51% plate color:100%
38	Face Detection	cf_retinaface_wider_360_640_1.11G_1.4	caffe	360*640	1.11G	wideface and FDDB	widerface and FDDB	0.9140@fp=100	0.8940@fp=100
39	Face Quality	cf_face-quality_80_60_61.68M_1.4	caffe	80*60	61.68M	private data	private data	12.5481 (L1 loss)	12.6863 (L1 loss)
40	Robot Instrument Segmentation	cf_FPN-resnet18_Endov_240_320_13.75G_1.4	caffe	240*320	13.75G	EndoVis'15	EndoVis'15	Dice=0.8050 Jaccard=0.7270	Dice=0.8010 Jaccard=0.7230
41	Pose Estimation	cf_hourglass_mpii_256_256_10.2G_1.4	caffe	256*256	10.20G	MPII Human Pose	MPII Human Pose	0.8718 (PCKh0.5)	0.8661 (PCKh0.5)
42	2D Detection	dk_tiny-yolov3_416_416_5.46G_1.4	darknet	416*416	5.46G	private data	private data	0.9739	0.9650
43	2D Detection	dk_yolov4_coco_416_416_60.1G_1.4	darknet	416*416	60.1G	coco	coco2014-5k	0.3950	0.3730
44	2D Detection	dk_yolov4_coco_416_416_0.36_38.2G_1.4	darknet	416*416	38.2G	coco	coco2014-5k	0.3810	0.3590
45	Classifiction	tf_inceptionresnetv2_imagenet_299_299_26.35G_1.4	tensorflow	299*299	26.35G	ImageNet Train	ImageNet Val	0.8037	0.7946
46	Classifiction	tf_inceptionv1_imagenet_224_224_3G_1.4	tensorflow	224*224	3G	ImageNet Train	ImageNet Val	0.6976	0.6794
47	Classifiction	tf_inceptionv3_imagenet_299_299_11.45G_1.4	tensorflow	299*299	11.45G	ImageNet Train	ImageNet Val	0.7798	0.7607
48	Classifiction	tf_inceptionv4_imagenet_299_299_24.55G_1.4	tensorflow	299*299	24.55G	ImageNet Train	ImageNet Val	0.8018	0.7928
49	Classifiction	tf_mobilenetv1_0.25_imagenet_128_128_27M_1.4	tensorflow	128*128	27.15M	ImageNet Train	ImageNet Val	0.4144	0.3464
50	Classifiction	tf_mobilenetv1_0.5_imagenet_160_160_150M_1.4	tensorflow	160*160	150.07M	ImageNet Train	ImageNet Val	0.5903	0.5195
51	Classifiction	tf_mobilenetv1_1.0_imagenet_224_224_1.14G_1.4	tensorflow	224*224	1.14G	ImageNet Train	ImageNet Val	0.7102	0.6779
52	Classifiction	tf_mobilenetv2_1.0_imagenet_224_224_602M_1.4	tensorflow	224*224	0.59G	ImageNet Train	ImageNet Val	0.7013	0.6767
53	Classifiction	tf_mobilenetv2_1.4_imagenet_224_224_1.16G_1.4	tensorflow	224*224	1.16G	ImageNet Train	ImageNet Val	0.7411	0.7194
54	Classifiction	tf_resnetv1_50_imagenet_224_224_6.97G_1.4	tensorflow	224*224	6.97G	ImageNet Train	ImageNet Val	0.7520	0.7478
55	Classifiction	tf_resnetv1_101_imagenet_224_224_14.4G_1.4	tensorflow	224*224	14.4G	ImageNet Train	ImageNet Val	0.7640	0.7560
56	Classifiction	tf_resnetv1_152_imagenet_224_224_21.83G_1.4	tensorflow	224*224	21.83G	ImageNet Train	ImageNet Val	0.7681	0.7463
57	Classifiction	tf_vgg16_imagenet_224_224_30.96G_1.4	tensorflow	224*224	30.96G	ImageNet Train	ImageNet Val	0.7089	0.7069
58	Classifiction	tf_vgg19_imagenet_224_224_39.28G_1.4	tensorflow	224*224	39.28G	ImageNet Train	ImageNet Val	0.7100	0.7026
59	Object Detection	tf_ssdmobilenetv1_coco_300_300_2.47G_1.4	tensorflow	300*300	2.47G	coco2017	coco2014 minival	0.2080	0.2100
60	Object Detection	tf_ssdmobilenetv2_coco_300_300_3.75G_1.4	tensorflow	300*300	3.75G	coco2017	coco2014 minival	0.2150	0.2110
61	Object Detection	tf_ssdresnet50v1_fpn_coco_640_640_178.4G_1.4	tensorflow	300*300	178.4G	coco2017	coco2014 minival	0.3010	0.2900
62	Object Detection	tf_yolov3_voc_416_416_65.63G_1.4	tensorflow	416*416	65.63G	voc07+12_trainval	voc07_test	0.7846	0.7744
63	Object Detection	tf_mlperf_resnet34_coco_1200_1200_433G_1.4	tensorflow	1200*1200	433G	coco2017	coco2017	0.2250	0.2150
64	Classifiction	tf_inceptionv2_imagenet_224_224_3.88G_1.4	tensorflow	224*224	3.88G	ImageNet Train	ImageNet Val	0.7399	07331
65	Classifiction	tf_resnetv2_50_imagenet_299_299_13.1G_1.4	tensorflow	299*299	13.1G	ImageNet Train	ImageNet Val	0.7559	0.7445
66	Classifiction	tf_resnetv2_101_imagenet_299_299_26.78G_1.4	tensorflow	299*299	26.78G	ImageNet Train	ImageNet Val	0.7695	0.7506
67	Classifiction	tf_resnetv2_152_imagenet_299_299_40.47G_1.4	tensorflow	299*299	40.47G	ImageNet Train	ImageNet Val	0.7779	0.7432
68	Object Detection	tf_ssdlite_mobilenetv2_coco_300_300_1.5G_1.4	tensorflow	300*300	1.5G	coco2017	coco2014 minival	0.2170	0.2090
69	Object Detection	tf_ssdinceptionv2_coco_300_300_9.62G_1.4	tensorflow	300*300	9.62G	coco2017	coco2014 minival	0.2390	0.2360
70	Segmentation	tf_mobilenetv2_cityscapes_1024_2048_132.74G_1.4	tensorflow	1024*2048	132.74G	Cityscapes	Cityscapes	0.6263	0.4578
71	Classification	tf_efficientnet-edgetpu-S_imagenet_224_224_4.72G_1.4	tensorflow	224*224	4.72G	ImageNet Train	ImageNet Val	0.7702/0.9377	0.7660/0.9337
72	Classification	tf_efficientnet-edgetpu-M_imagenet_240_240_7.34G_1.4	tensorflow	240*240	7.34G	ImageNet Train	ImageNet Val	0.7862/0.9440	0.7798/0.9406
73	Classification	tf_efficientnet-edgetpu-L_imagenet_300_300_19.36G_1.4	tensorflow	300*300	19.36G	ImageNet Train	ImageNet Val	0.8026/0.9514	0.7996/0.9491
74	Classifiction	tf_mlperf_resnet50_imagenet_224_224_8.19G_1.4	tensorflow	224*224	8.19G	ImageNet Train	ImageNet Val	0.7652	0.7606
75	General Detection	tf_refinedet_VOC_320_320_81.9G_1.4	tensorflow	320*320	81.9G	voc07+12_trainval	voc07_test	0.8015	0.7999
76	Classifiction	tf_mobilenetEdge1.0_imagenet_224_224_990M_1.4	tensorflow	224*224	990M	ImageNet Train	ImageNet Val	0.7227	0.6775
77	Classifiction	tf_mobilenetEdge0.75_imagenet_224_224_624M_1.4	tensorflow	224*224	624M	ImageNet Train	ImageNet Val	0.7201	0.6489
78	Medical Detection	tf_RefineDet-Medical_EDD_320_320_9.83G_1.4	tensorflow	320*320	9.83G	EDD	EDD	0.7839	0.8014
79	Super Resolution	tf_rcan_DIV2K_360_640_0.98_86.95G_1.4	tensorflow	360*640	86.95G	DIV2K	DIV2K	Set5 Y_PSNR : 37.6402 SSIM : 0.9592	Set5 Y_PSNR : 37.2495 SSIM : 0.9556
80	Classifiction	tf2_resnet50_imagenet_224_224_7.76G_1.4	tensorflow	224*224	7.76G	ImageNet Train	ImageNet Val	0.7513	0.7423
81	Classifiction	tf2_mobilenetv1_imagenet_224_224_1.15G_1.4	tensorflow	224*224	1.15G	ImageNet Train	ImageNet Val	0.7005	0.5603
82	Classifiction	tf2_inceptionv3_imagenet_299_299_11.5G_1.4	tensorflow	299*299	11.5G	ImageNet Train	ImageNet Val	0.7753	0.7694
83	Medical Segmentation	tf2_2d-unet_nuclei_128_128_5.31G_1.4	tensorflow	128*128	5.31G	Nuclei Cell	Nuclei Cell	0.3968	0.3968
84	Semantic Segmentation	tf2_erfnet_cityscapes_512_1024_54G_1.4	tensorflow	512*1024	54G	Cityscapes	Cityscapes	0.5298	0.5167
85	Classifiction	tf2_efficientnet-b0_imagenet_224_224_0.36G_1.4	tensorflow	224*224	0.36G	ImageNet Train	ImageNet Val	0.7690/0.9320	0.7515/0.9273
86	Segmentation	pt_ENet_cityscapes_512_1024_8.6G_1.4	pytorch	512*1024	8.6G	Cityscapes	Cityscapes	0.6442	0.6315
87	Segmentation	pt_SemanticFPN-resnet18_cityscapes_256_512_10G_1.4	pytorch	256*512	10G	Cityscapes	Cityscapes	0.6290	0.6230
88	Face Recognition	pt_facerec-resnet20_mixed_112_96_3.5G_1.4	pytorch	112*96	3.5G	mixed	mixed	0.9955	0.9947
89	Face Quality	pt_face-quality_80_60_61.68M_1.4	pytorch	80*60	61.68M	private data	private data	0.1233	0.1258
90	Multi Task	pt_MT-resnet18_mixed_320_512_13.65G_1.4	pytorch	320*512	13.65G	mixed	mixed	mAP:0.3951 mIOU:0.4403	mAP:0.3841 mIOU:0.4271
91	Face ReID	pt_facereid-large_96_96_515M_1.4	pytorch	96*96	515M	private data	private data	mAP:0.7940 Rank1:0.9550	mAP:0.7900 Rank1:0.9530
92	Face ReID	pt_facereid-small_80_80_90M_1.4	pytorch	80*80	90M	private data	private data	mAP:0.5600 Rank1:0.8650	mAP:0.5590 Rank1:0.8650
93	Re-identification	pt_personreid-res50_market1501_256_128_5.4G_1.4	pytorch	256*128	4.2G	market1501	market1501	mAP:0.8540 Rank1:0.9410	mAP:0.8380 Rank1:0.9370
94	Re-identification	pt_personreid-res18_market1501_176_80_1.1G_1.4	pytorch	176*80	1.1G	market1501	market1501	mAP:0.7530 Rank1:0.8980	mAP:0.7460 Rank1:0.8930
95	3D Detection	pt_pointpillars_kitti_12000_100_10.8G_1.4	pytorch	12000*100*4	10.8G	kitti	kitti	Car 3D [email protected](easy, moderate, hard) 90.79, 89.66, 88.78	Car 3D [email protected](easy, moderate, hard) 90.75, 87.04, 83.44
96	3D point cloud Segmentation	pt_salsanext_semantic-kitti_64_2048_0.6_20.4G_1.4	pytorch	64*2048	20.4G	semantic-kitti	semantic-kitti	Acc avg 0.8860 IoU avg 0.5100	Acc avg 0.8350 IoU avg 0.4540
97	Covid-19 Segmentation	pt_FPN-resnet18_covid19-seg_352_352_22.7G_1.4	pytorch	352*352	22.7G	COVID19-seg	COVID19-seg	2-classes Dice:0.8588 3-classes mIoU:0.5989	2-classes Dice:0.8547 3-classes mIoU:0.5957
98	CT lung Segmentation	pt_unet_chaos-CT_512_512_23.3G_1.4	pytorch	512*512	23.3G	Chaos-CT	Chaos-CT	Dice:0.9758	Dice:0.9747
99	surround-view 3D Detection	pt_pointpillars_nuscenes_40000_64_108G_1.4	pytorch	40000*64*5	108G	NuScenes	NuScenes	mAP: 42.2 NDS: 55.1	mAP: 40.5 NDS: 53.0
100	3D Segmentation	pt_salsanextv2_semantic-kitti_64_2048_0.75_32G_1.4	pytorch	64*2048*5	32G	semantic-kitti	semantic-kitti	mIou: 54.2%	mIou: 54.2%(QAT)
101	4D radar based 3D Detection	pt_centerpoint_astyx_2560_40_54G_1.4	pytorch	2560*40*4	54G	Astyx 4D radar	Astyx 4D radar	BEV [email protected]: 32.843D [email protected]: 28.27	BEV [email protected]: 33.823D [email protected]: 18.54(QAT)
102	Image-lidar sensor fusion	pt_pointpainting_nuscenes_126G_1.4	pytorch	semanticfpn:320*576 pointpillars:40000*64*16	semanticfpn:14G pointpillars:112G	semanticfpn:NuImages v1.0 pointpillars:NuScenes v1.0	semanticfpn:NuImages v1.0 pointpillars:NuScenes v1.0	mIoU: 69.1% mAP: 51.8 NDS: 58.7	mIoU: 68.6% mAP: 50.4 NDS: 56.4(QAT)
103	Multi Task	pt_multitaskv3_mixed_320_512_25.44G_1.4	pytorch	320*512	25.44G	mixed	mixed	mAP:51.2 mIOU:58.14 Drivable mIOU: 82.57 Lane IOU:43.71 Silog: 8.78	mAP:50.9 mIOU:57.52 Drivable mIOU: 82.30 Lane IOU:44.01 Silog: 9.32
104	Depth Estimation	pt_fadnet_sceneflow_576_960_359G_1.4	pytorch	576*960	359G	Scene Flow	Scene Flow	EPE: 0.926	EPE: 1.169
105	RGB-D Segmentation	pt_sa-gate_NYUv2_360_360_178G_1.4	pytorch	360*360	178G	NYUv2	NYUv2	miou: 50.41%	miou: 49.61%
106	Crowd Counting	pt_BCC_shanghaitech_800_1000_268.9G_1.4	pytorch	800*1000	268.9G	shanghaitech A	shanghaitech A	MAE: 65.83 MSE: 111.75	MAE: 67.60 MSE: 117.36(QAT)
107	Production Recognition	pt_pmg_rp2k_224_224_2.28G_1.4	pytorch	224*224	2.28G	RP2K	RP2K	0.9640	0.9618
108	Segmentation	pt_SemanticFPN-mobilenetv2_cityscapes_512_1024_5.4G_1.4	pytorch	512*1024	5.4G	Cityscapes	Cityscapes	0.6870	0.6820

Naming Rules

Model name: F_M_(D)_H_W_(P)_C_V

• F specifies training framework: cf is Caffe, tf is Tensorflow, tf2 is Tensorflow 2, dk is Darknet, pt is PyTorch
• M specifies the model
• D specifies the dataset. It is optional depending on whether the dataset is public or private
• H specifies the height of input data
• W specifies the width of input data
• P specifies the pruning ratio, it means how much computation is reduced. It is optional depending on whether the model is pruned or not
• C specifies the computation of the model: how many Gops per image
• V specifies the version of Vitis AI

For example, cf_refinedet_coco_360_480_0.8_25G_1.4 is a RefineDet model trained with Caffe using COCO dataset, input data size is 360*480, 80% pruned, the computation per image is 25Gops and Vitis AI version is 1.4.

caffe-xilinx

This is a custom distribution of caffe. Please use caffe-xilinx to test and finetune the caffe models listed in this page. If you use the Vitis AI docker environment to evaluate and develop the caffe models, you will not need to perform this step, as caffe-xilinx has been pre-compiled in docker.

Model Download

Please visit model-list in this page. You will get downloadlink and MD5 of all the released models, including pre-compiled models running on different platforms.

Automated Download Script

With downloader.py, you could quickly find the model you are interested in and specify a version to download it immediately. Please make sure that downloader.py and model-list folder are at the same level directory.

python3  downloader.py  

Step1: You need input framework and model name keyword. Use space divide.

tf: tensorflow1.x, tf2: tensorflow2.x, pt: pytorch, cf: caffe, dk: darknet

Step2: Select the specified model based on standard name.

Step3: Select the specified hardware platform for your slected model.

For example, after running downloader.py and input cf densebox then you will see the alternatives such as:

0:  cf_densebox_wider_320_320_0.49G_1.4
1:  cf_densebox_wider_360_640_1.11G_1.4 

After you input the num: 0, you will see the alternatives such as:

0:  cf_densebox_wider_320_320_0.49G_1.4    GPU
1:  densebox_320_320    ZCU102 & ZCU104 & KV260
2:  densebox_320_320    VCK190
3:  densebox_320_320    U50
4:  densebox_320_320    U50lv
5:  densebox_320_320    U50-V3ME
......

Then you could choose and input the number, the specified version of model will be automatically downloaded to the current directory.

Model Directory Structure

Download and extract the model archive to your working area on the local hard disk. For details on the various models, download link and MD5 checksum for the zip file of each model, see model-list.

Caffe Model Directory Structure

For a caffe model, you should see the following directory structure:

├── code                            # Contains test , training and quantization scripts.
│                                     
│                                   
├── readme.md                       # Contains the environment requirements, data preprocess and model information.
│                                     Refer this to know that how to test and train the model with scripts.
│                                        
├── data                            # Contains the dataset that used for model test and training.
│                                     When test or training scripts run successfully, dataset will be automatically placed in it.
│                                     In some cases, you may also need to manually place your own dataset here.
│                                                      
├── quantized  
│    │    
│    ├── deploy.caffemodel                # Quantized weights, the output of vai_q_caffe without modification.
│    ├── deploy.prototxt                  # Quantized prototxt, the output of vai_q_caffe without modification.
│    ├── quantize_test.prototxt           # Used to run evaluation with quantize_train_test.caffemodel. 
│    │                                      Some models don't have this file if they are converted from Darknet (Yolov2, Yolov3),   
│    │                                      Pytorch (ReID) or there is no Caffe Test (Densebox). 
│    │
│    ├── quantize_train_test.caffemodel   # Quantized weights can be used for quantized-point training and evaluation.       
│    └── quantize_train_test.prototxt     # Used for quantized-point training and testing with quantize_train_test.caffemodel 
│                                           on GPU when datalayer modified to user's data path.         
│                                                 
└── float                           
     ├── trainval.caffemodel              # Trained float-point weights.
     ├── test.prototxt                    # Used to run evaluation with python test codes released in near future.    
     └── trainval.prorotxt                # Used for training and testing with caffe train/test command 
                                            when datalayer modified to user's data path.Some models don't
                                            have this file if they are converted from Darknet (Yolov2, Yolov3),
                                            Pytorch (ReID) or there is no Caffe Test (Densebox).              

Tensorflow Model Directory Structure

For a Tensorflow model, you should see the following directory structure:

├── code                            # Contains test code which can run demo and evaluate model performance. 
│                          
│
├── readme.md                       # Contains the environment requirements, data preprocess and model information.
│                                     Refer this to know that how to test the model with scripts.
│
├── data                            # Contains the dataset that used for model test and training.
│                                     When test or training scripts run successfully, dataset will be automatically placed in it.
│
├── quantized                          
│   └── quantize_eval_model.pb      # Quantized model for evaluation.
│
└── float                             
    └── frozen.pb                   # Float-point frozen model, the input to the `vai_q_tensorflow`.
                                      The pb name of different models may be different.

Pytorch Model Directory Structure

For a Pytorch model, you should see the following directory structure:

├── code                            # Contains test and training code.  
│                                                        
│                                   
├── readme.md                       # Contains the environment requirements, data preprocess and model information.
│                                     Refer this to know that how to test and train the model with scripts.
│                                        
├── data                            # Contains the dataset that used for model test and training.
│                                     When test or training scripts run successfully, dataset will be automatically placed in it.
│
├── qat                             # Contains the QAT(Quantization Aware Training) results. 
│                                     The accuracy of QAT result is better than direct quantization. 
│                                     Some models but not all supported QAT, and only these models have qat folder. 
│                                         
├── quantized                          
│   ├── _int.pth                    # Quantized model.
│   ├── quant_info.json             # Quantization steps of tensors got. Please keep it for evaluation of quantized model.
│   ├── _int.py                     # Converted vai_q_pytorch format model.
│   └── _int.xmodel                 # Deployed model. The name of different models may be different.
│                                     For some models that support QAT you could find better quantization results in 'qat' folder. 
│
│
└── float                           
    └── _int.pth                    # Trained float-point model. The pth name of different models may be different.
                                      Path and model name in test scripts could be modified according to actual situation.

Note: For more information on vai_q_caffe, vai_q_tensorflow and vai_q_pytorch, please see the Vitis AI User Guide.

Model Performance

All the models in the Model Zoo have been deployed on Xilinx hardware with Vitis AI and Vitis AI Library. The performance number including end-to-end throughput and latency for each model on various boards with different DPU configurations are listed in the following sections.

For more information about DPU, see DPU IP Product Guide.

Note: The model performance number listed in the following sections is generated with Vitis AI v1.4 and Vitis AI Lirary v1.4. For different boards, different DPU configurations are used. Vitis AI and Vitis AI Library can be downloaded for free from Vitis AI Github and Vitis AI Library Github. We will continue to improve the performance with Vitis AI. The performance number reported here is subject to change in the near future.

Performance on ZCU102 (0432055-05)

Measured with Vitis AI 1.4 and Vitis AI Library 1.4

Click here to view details

The following table lists the performance number including end-to-end throughput and latency for each model on the ZCU102 (0432055-05) board with a 3 * B4096 @ 281MHz DPU configuration:

No.	Model	Name	E2E latency (ms) Thread Num =1	E2E throughput (fps) Single Thread	E2E throughput (fps) Multi Thread
1	resnet50	cf_resnet50_imagenet_224_224_7.7G	13.75	72.7	173.6
2	resnet18	cf_resnet18_imagenet_224_224_3.65G	5.46	183.1	478
3	Inception_v1	cf_inceptionv1_imagenet_224_224_3.16G	5.57	179.5	469.8
4	Inception_v2	cf_inceptionv2_imagenet_224_224_4G	7.6	131.6	303.2
5	Inception_v3	cf_inceptionv3_imagenet_299_299_11.4G	16.99	58.8	135.9
6	Inception_v4	cf_inceptionv4_imagenet_299_299_24.5G	34.95	28.6	68.5
7	Mobilenet_v2	cf_mobilenetv2_imagenet_224_224_0.59G	4	250	734
8	SqueezeNet	cf_squeezenet_imagenet_227_227_0.76G	3.69	270.8	1069
9	ssd_pedestrian_pruned_0_97	cf_ssdpedestrian_coco_360_640_0.97_5.9G	13.14	76.1	278.6
10	refinedet_baseline	cf_refinedet_coco_360_480_123G	117.37	8.5	24.7
11	refinedet_pruned_0_8	cf_refinedet_coco_360_480_0.8_25G	30.74	32.5	100.1
12	refinedet_pruned_0_92	cf_refinedet_coco_360_480_0.92_10.10G	15.9	62.9	201.3
13	refinedet_pruned_0_96	cf_refinedet_coco_360_480_0.96_5.08G	11.46	87.2	288
14	ssd_adas_pruned_0_95	cf_ssdadas_bdd_360_480_0.95_6.3G	11.46	87.2	297.3
15	ssd_traffic_pruned_0_9	cf_ssdtraffic_360_480_0.9_11.6G	18.22	54.9	200.1
16	VPGnet_pruned_0_99	cf_VPGnet_caltechlane_480_640_0.99_2.5G	10.59	94.4	351.3
17	ssd_mobilenet_v2	cf_ssdmobilenetv2_bdd_360_480_6.57G	24.99	40	117.6
18	FPN	cf_fpn_cityscapes_256_512_8.9G	29.03	34.4	151.9
19	SP_net	cf_SPnet_aichallenger_224_128_0.54G	1.94	515.1	1588.7
20	Openpose_pruned_0_3	cf_openpose_aichallenger_368_368_0.3_189.7G	265.03	3.8	15.2
21	densebox_320_320	cf_densebox_wider_320_320_0.49G	2.38	420.2	1630
22	densebox_640_360	cf_densebox_wider_360_640_1.11G	4.78	209	812.9
23	face_landmark	cf_landmark_celeba_96_72_0.14G	1.19	841.6	1547.6
24	reid	cf_reid_market1501_160_80_0.95G	2.85	350.5	692.7
25	multi_task	cf_multitask_bdd_288_512_14.8G	26.35	37.9	127
26	yolov3_bdd	dk_yolov3_bdd_288_512_53.7G	80.56	12.4	33
27	yolov3_adas_pruned_0_9	dk_yolov3_cityscapes_256_512_0.9_5.46G	11.02	90.7	263.4
28	yolov3_voc	dk_yolov3_voc_416_416_65.42G	78.5	12.7	33.5
29	yolov2_voc	dk_yolov2_voc_448_448_34G	37.88	26.4	70
30	yolov2_voc_pruned_0_66	dk_yolov2_voc_448_448_0.66_11.56G	15.9	62.9	191.6
31	yolov2_voc_pruned_0_71	dk_yolov2_voc_448_448_0.71_9.86G	13.81	72.4	224
32	yolov2_voc_pruned_0_77	dk_yolov2_voc_448_448_0.77_7.82G	11.83	84.5	269.3
33	ResNet20-face	cf_facerec-resnet20_112_96_3.5G	6.11	163.5	329.8
34	ResNet64-face	cf_facerec-resnet64_112_96_11G	14.13	70.7	176.8
35	FPN_Res18_segmentation	cf_FPN-resnet18_EDD_320_320_45.3G	81.65	12.2	45.8
36	plate detection	cf_plate-detection_320_320_0.49G	1.97	506.2	2009.4
37	plate recognition	cf_plate-recognition_96_288_1.75G	6.66	150.2	494.5
38	retinaface	cf_retinaface_wider_360_640_1.11G	7.94	125.9	542.1
39	face_quality	cf_face-quality_80_60_61.68M	0.47	2118.7	7244.3
40	FPN-R18(light-weight)	cf_FPN-resnet18_Endov_240_320_13.75G	27.52	36.3	155.9
41	Hourglass	cf_hourglass_mpii_256_256_10.2G	54.76	18.3	73.5
42	tiny-yolov3	dk_tiny-yolov3_416_416_5.46G	8.51	117.5	393
43	yolov4	dk_yolov4_coco_416_416_60.1G	75.45	13.2	33.8
44	pruned_yolov4	dk_yolov4_coco_416_416_0.36_38.2G	55.4	18	45.5
45	Inception_resnet_v2	tf_inceptionresnetv2_imagenet_299_299_26.35G	44.67	22.4	50.8
46	Inception_v1	tf_inceptionv1_imagenet_224_224_3G	5.47	182.8	468.9
47	Inception_v3	tf_inceptionv3_imagenet_299_299_11.45G	17.03	58.7	134.8
48	Inception_v4	tf_inceptionv4_imagenet_299_299_24.55G	34.97	28.6	68.5
49	Mobilenet_v1	tf_mobilenetv1_0.25_imagenet_128_128_27.15M	0.89	1122.8	4147.4
50	Mobilenet_v1	tf_mobilenetv1_0.5_imagenet_160_160_150.07M	1.4	712.3	2657.2
51	Mobilenet_v1	tf_mobilenetv1_1.0_imagenet_224_224_1.14G	3.37	296.8	940.9
52	Mobilenet_v2	tf_mobilenetv2_1.0_imagenet_224_224_0.59G	4.08	244.7	694.9
53	Mobilenet_v2	tf_mobilenetv2_1.4_imagenet_224_224_1.16G	5.57	179.5	467.9
54	resnet_v1_50	tf_resnetv1_50_imagenet_224_224_6.97G	12.56	79.5	185.6
55	resnet_v1_101	tf_resnetv1_101_imagenet_224_224_14.4G	23.48	42.6	106.4
56	resnet_v1_152	tf_resnetv1_152_imagenet_224_224_21.83G	34.29	29.1	74.1
57	vgg_16	tf_vgg16_imagenet_224_224_30.96G	49.75	20.1	41.1
58	vgg_19	tf_vgg19_imagenet_224_224_39.28G	57.68	17.3	36.6
59	ssd_mobilenet_v1	tf_ssdmobilenetv1_coco_300_300_2.47G	9.39	106.5	330.5
60	ssd_mobilenet_v2	tf_ssdmobilenetv2_coco_300_300_3.75G	12.67	78.9	212.4
61	ssd_resnet_50_v1_fpn	tf_ssdresnet50v1_fpn_coco_640_640_178.4G	346.97	2.9	5.1
62	yolov3_voc	tf_yolov3_voc_416_416_65.63G	75.63	13.2	34.5
63	mlperf_ssd_resnet34	tf_mlperf_resnet34_coco_1200_1200_433G	559.75	1.8	7.1
64	Inception_v2	tf_inceptionv2_imagenet_224_224_3.88G	10.97	91.1	230
65	resnet_v2_50	tf_resnetv2_50_imagenet_299_299_13.1G	28.49	35.1	95.5
66	resnet_v2_101	tf_resnetv2_101_imagenet_299_299_26.78G	48.77	20.5	54.4
67	resnet_v2_152	tf_resnetv2_152_imagenet_299_299_40.47G	69.03	14.5	37.5
68	ssdlite_mobilenetv2	tf_ssdlite_mobilenetv2_coco_300_300_1.5G	9.92	100.7	305.2
69	ssd_inceptionv2	tf_ssdinceptionv2_coco_300_300_9.62G	28.58	35	101.5
70	Mobilenet_v2	tf_mobilenetv2_cityscapes_1024_2048_132.74G	620.95	1.6	4.6
71	efficientnet-edgetpu-S	tf_efficientnet-edgetpu-S_imagenet_224_224_4.72G	10.54	94.9	260.9
72	efficientnet-edgetpu-M	tf_efficientnet-edgetpu-M_imagenet_240_240_7.34G	14.53	68.8	181.7
73	efficientnet-edgetpu-L	tf_efficientnet-edgetpu-L_imagenet_300_300_19.36G	35.77	27.9	78
74	mlperf_resnet50	tf_mlperf_resnet50_imagenet_224_224_8.19G	14.04	71.2	168.5
75	refinedet	tf_refinedet_VOC_320_320_81.9G	90.56	11	34.4
76	mobilenet_edge_1.0	tf_mobilenetEdge1.0_imagenet_224_224_990M	5.04	198.2	547.3
77	mobilenet_edge_0.75	tf_mobilenetEdge0.75_imagenet_224_224_624M	4.21	237.5	706.7
78	refinedet_medical	tf_RefineDet-Medical_EDD_320_320_9.83G	15.13	66.1	230.4
79	pruned_rcan	tf_rcan_DIV2K_360_640_0.98_86.95G	131.87	7.6	17.1
80	resnet50	tf2_resnet50_imagenet_224_224_7.76G	14.15	70.6	171.2
81	Mobilenet_v1	tf2_mobilenetv1_imagenet_224_224_1.15G	3.42	292.5	930.9
82	Inception_v3	tf2_inceptionv3_imagenet_299_299_11.5G	17.15	58.3	136.6
83	2d-unet	tf2_2d-unet_nuclei_128_128_5.31G	6.46	154.8	395.1
84	ERFNet	tf2_erfnet_cityscapes_512_1024_54G	141.25	7.1	24
85	efficientnet-b0	tf2_efficientnet-b0_imagenet_224_224_0.36G	-	-	-
86	ENet	pt_ENet_cityscapes_512_1024_8.6G	108.07	9.3	36.9
87	SemanticFPN	pt_SemanticFPN_cityscapes_256_512_10G	29.34	34.1	161.5
88	ResNet20-face	pt_facerec-resnet20_mixed_112_96_3.5G	6.12	163.3	329.7
89	face quality	pt_face-quality_80_60_61.68M	0.48	2098	7196.2
90	multi_task_v2	pt_MT-resnet18_mixed_320_512_13.65G	32	31.2	101.8
91	face_reid_large	pt_facereid-large_96_96_515M	1.19	841.3	2205.7
92	face_reid_small	pt_facereid-small_80_80_90M	0.54	1861.8	5973.2
93	person_reid	pt_personreid-res50_market1501_256_128_5.4G	10.33	96.8	228.2
94	person_reid	pt_personreid-res18_market1501_176_80_1.1G	2.86	349.2	684.1
95	pointpillars	pt_pointpillars_kitti_12000_100_10.8G	50.34	19.8	50
96	salsanext	pt_salsanext_semantic-kitti_64_2048_0.6_20.4G	181.45	5.5	21.1
97	FPN-R18 (light-weight)	pt_FPN-resnet18_covid19-seg_352_352_22.7G	27.93	35.8	106.1
98	2d-unet	pt_unet_chaos-CT_512_512_23.3G	45.15	22.1	69.7
99	surround-view pointpillars	pt_pointpillars_nuscenes_40000_64_108G	440.5	2.3	9.9
100	salsanext_v2	pt_salsanextv2_semantic-kitti_64_2048_0.75_32G	246.30	4.1	11.1
101	centerpoint	pt_centerpoint_astyx_2560_40_54G	62.45	16	48
102	pointpainting	pt_pointpainting_nuscenes_126G	779.83	1.3	4.4
103	multi_task_v3	pt_multitaskv3_mixed_320_512_25.44G	60.01	16.6	60.8
104	FADnet	pt_fadnet_sceneflow_576_960_359G	-	-	-
105	SA-gate	pt_sa-gate_NYUv2_360_360_178G	-	-	-
106	Bayesian Crowd Counting	pt_BCC_shanghaitech_800_1000_268.9G	304.73	3.3	10.7
107	PMG	pt_pmg_rp2k_224_224_2.28G	6.91	144.7	363.4
108	SemanticFPN-mobilenetv2	pt_SemanticFPN-mobilenetv2_cityscapes_512_1024_5.4G	98.54	10.1	54
-	Inception_v3	torchvision_inception_v3	16.99	58.8	136.1
-	SqueezeNet	torchvision_squeezenet	4.51	221.6	816.8
-	resnet50	torchvision_resnet50	14.54	68.7	166.5

Performance on ZCU104

Measured with Vitis AI 1.4 and Vitis AI Library 1.4

Click here to view details

The following table lists the performance number including end-to-end throughput and latency for each model on the ZCU104 board with a 2 * B4096 @ 300MHz DPU configuration:

No.	Model	Name	E2E latency (ms) Thread Num =1	E2E throughput (fps) Single Thread	E2E throughput (fps) Multi Thread
1	resnet50	cf_resnet50_imagenet_224_224_7.7G	12.89	77.5	151.8
2	resnet18	cf_resnet18_imagenet_224_224_3.65G	5.16	193.8	416.9
3	Inception_v1	cf_inceptionv1_imagenet_224_224_3.16G	5.27	189.8	405.7
4	Inception_v2	cf_inceptionv2_imagenet_224_224_4G	7.16	139.7	278.6
5	Inception_v3	cf_inceptionv3_imagenet_299_299_11.4G	16.01	62.4	122
6	Inception_v4	cf_inceptionv4_imagenet_299_299_24.5G	32.81	30.5	59
7	Mobilenet_v2	cf_mobilenetv2_imagenet_224_224_0.59G	3.8	263	610.6
8	SqueezeNet	cf_squeezenet_imagenet_227_227_0.76G	3.62	276.2	978.3
9	ssd_pedestrian_pruned_0_97	cf_ssdpedestrian_coco_360_640_0.97_5.9G	12.57	79.5	215.6
10	refinedet_baseline	cf_refinedet_coco_360_480_123G	110.18	9.1	18.4
11	refinedet_pruned_0_8	cf_refinedet_coco_360_480_0.8_25G	28.99	34.5	74.4
12	refinedet_pruned_0_92	cf_refinedet_coco_360_480_0.92_10.10G	15.08	66.3	155
13	refinedet_pruned_0_96	cf_refinedet_coco_360_480_0.96_5.08G	10.9	91.7	227
14	ssd_adas_pruned_0_95	cf_ssdadas_bdd_360_480_0.95_6.3G	10.94	91.3	238.4
15	ssd_traffic_pruned_0_9	cf_ssdtraffic_360_480_0.9_11.6G	17.38	57.5	150.4
16	VPGnet_pruned_0_99	cf_VPGnet_caltechlane_480_640_0.99_2.5G	10.2	98	300.7
17	ssd_mobilenet_v2	cf_ssdmobilenetv2_bdd_360_480_6.57G	39.67	25.2	106.2
18	FPN	cf_fpn_cityscapes_256_512_8.9G	28.33	35.3	146.4
19	SP_net	cf_SPnet_aichallenger_224_128_0.54G	1.86	538	1330.7
20	Openpose_pruned_0_3	cf_openpose_aichallenger_368_368_0.3_189.7G	252.51	4	11
21	densebox_320_320	cf_densebox_wider_320_320_0.49G	2.32	430.1	1560
22	densebox_640_360	cf_densebox_wider_360_640_1.11G	4.67	214.3	754.4
23	face_landmark	cf_landmark_celeba_96_72_0.14G	1.12	888.5	1591.8
24	reid	cf_reid_market1501_160_80_0.95G	2.68	372.7	705.4
25	multi_task	cf_multitask_bdd_288_512_14.8G	25.21	39.6	109.1
26	yolov3_bdd	dk_yolov3_bdd_288_512_53.7G	75.66	13.2	26.8
27	yolov3_adas_pruned_0_9	dk_yolov3_cityscapes_256_512_0.9_5.46G	10.53	94.9	233.8
28	yolov3_voc	dk_yolov3_voc_416_416_65.42G	73.71	13.6	27.3
29	yolov2_voc	dk_yolov2_voc_448_448_34G	35.64	28	57.5
30	yolov2_voc_pruned_0_66	dk_yolov2_voc_448_448_0.66_11.56G	15.03	66.5	153.3
31	yolov2_voc_pruned_0_71	dk_yolov2_voc_448_448_0.71_9.86G	13.09	76.4	180.9
32	yolov2_voc_pruned_0_77	dk_yolov2_voc_448_448_0.77_7.82G	11.23	89	217.3
33	ResNet20-face	cf_facerec-resnet20_112_96_3.5G	5.74	174.2	309.5
34	ResNet64-face	cf_facerec-resnet64_112_96_11G	13.26	75.4	145
35	FPN_Res18_segmentation	cf_FPN-resnet18_EDD_320_320_45.3G	77.75	12.9	33.5
36	plate detection	cf_plate-detection_320_320_0.49G	1.94	514.5	1926.1
37	plate recognition	cf_plate-recognition_96_288_1.75G	6.13	162.9	440
38	retinaface	cf_retinaface_wider_360_640_1.11G	7.72	129.4	481.2
39	face_quality	cf_face-quality_80_60_61.68M	0.46	2172.8	6828.2
40	FPN-R18(light-weight)	cf_FPN-resnet18_Endov_240_320_13.75G	26.62	37.6	129.7
41	Hourglass	cf_hourglass_mpii_256_256_10.2G	53.69	18.6	73.2
42	tiny-yolov3	dk_tiny-yolov3_416_416_5.46G	8.17	122.3	329.5
43	yolov4	dk_yolov4_coco_416_416_60.1G	71.04	14.1	28.5
44	pruned_yolov4	dk_yolov4_coco_416_416_0.36_38.2G	52.12	19.2	39
45	Inception_resnet_v2	tf_inceptionresnetv2_imagenet_299_299_26.35G	41.73	23.9	45.1
46	Inception_v1	tf_inceptionv1_imagenet_224_224_3G	5.17	193.3	412.1
47	Inception_v3	tf_inceptionv3_imagenet_299_299_11.45G	16.04	62.3	121.6
48	Inception_v4	tf_inceptionv4_imagenet_299_299_24.55G	32.82	30.4	59
49	Mobilenet_v1	tf_mobilenetv1_0.25_imagenet_128_128_27.15M	0.87	1151.3	4155.9
50	Mobilenet_v1	tf_mobilenetv1_0.5_imagenet_160_160_150.07M	1.35	740.4	2264.2
51	Mobilenet_v1	tf_mobilenetv1_1.0_imagenet_224_224_1.14G	3.21	311.8	769.7
52	Mobilenet_v2	tf_mobilenetv2_1.0_imagenet_224_224_0.59G	3.87	258.3	593.2
53	Mobilenet_v2	tf_mobilenetv2_1.4_imagenet_224_224_1.16G	5.25	190.3	406.1
54	resnet_v1_50	tf_resnetv1_50_imagenet_224_224_6.97G	11.78	84.8	165.2
55	resnet_v1_101	tf_resnetv1_101_imagenet_224_224_14.4G	22	45.4	89
56	resnet_v1_152	tf_resnetv1_152_imagenet_224_224_21.83G	32.13	31.1	60.8
57	vgg_16	tf_vgg16_imagenet_224_224_30.96G	46.68	21.4	37.1
58	vgg_19	tf_vgg19_imagenet_224_224_39.28G	54.07	18.5	32.7
59	ssd_mobilenet_v1	tf_ssdmobilenetv1_coco_300_300_2.47G	8.92	112.1	303.5
60	ssd_mobilenet_v2	tf_ssdmobilenetv2_coco_300_300_3.75G	12.04	83	194.1
61	ssd_resnet_50_v1_fpn	tf_ssdresnet50v1_fpn_coco_640_640_178.4G	346.53	2.9	5.2
62	yolov3_voc	tf_yolov3_voc_416_416_65.63G	71.02	14.1	28.2
63	mlperf_ssd_resnet34	tf_mlperf_resnet34_coco_1200_1200_433G	533.66	1.9	5.2
64	Inception_v2	tf_inceptionv2_imagenet_224_224_3.88G	10.31	97	195.2
65	resnet_v2_50	tf_resnetv2_50_imagenet_299_299_13.1G	26.86	37.2	84.3
66	resnet_v2_101	tf_resnetv2_101_imagenet_299_299_26.78G	45.78	21.8	46.1
67	resnet_v2_152	tf_resnetv2_152_imagenet_299_299_40.47G	64.68	15.4	31.5
68	ssdlite_mobilenetv2	tf_ssdlite_mobilenetv2_coco_300_300_1.5G	9.45	105.8	273.1
69	ssd_inceptionv2	tf_ssdinceptionv2_coco_300_300_9.62G	24.72	40.4	86.8
70	Mobilenet_v2	tf_mobilenetv2_cityscapes_1024_2048_132.74G	596.67	1.7	4.7
71	efficientnet-edgetpu-S	tf_efficientnet-edgetpu-S_imagenet_224_224_4.72G	9.9	101	208.4
72	efficientnet-edgetpu-M	tf_efficientnet-edgetpu-M_imagenet_240_240_7.34G	13.62	73.4	148
73	efficientnet-edgetpu-L	tf_efficientnet-edgetpu-L_imagenet_300_300_19.36G	33.73	29.6	63.1
74	mlperf_resnet50	tf_mlperf_resnet50_imagenet_224_224_8.19G	13.17	75.9	148.2
75	refinedet	tf_refinedet_VOC_320_320_81.9G	92.92	10.8	25.8
76	mobilenet_edge_1.0	tf_mobilenetEdge1.0_imagenet_224_224_990M	4.77	209.5	463.5
77	mobilenet_edge_0.75	tf_mobilenetEdge0.75_imagenet_224_224_624M	4	249.9	581.4
78	refinedet_medical	tf_RefineDet-Medical_EDD_320_320_9.83G	14.38	69.5	169.8
79	pruned_rcan	tf_rcan_DIV2K_360_640_0.98_86.95G	124.01	8.1	15.3
80	resnet50	tf2_resnet50_imagenet_224_224_7.76G	13.28	75.3	148
81	Mobilenet_v1	tf2_mobilenetv1_imagenet_224_224_1.15G	3.25	307.4	755.9
82	Inception_v3	tf2_inceptionv3_imagenet_299_299_11.5G	16.16	61.8	121.8
83	2d-unet	tf2_2d-unet_nuclei_128_128_5.31G	6.1	163.9	342.5
84	ERFNet	tf2_erfnet_cityscapes_512_1024_54G	136.94	7.3	25.4
85	efficientnet-b0	tf2_efficientnet-b0_imagenet_224_224_0.36G	-	-	-
86	ENet	pt_ENet_cityscapes_512_1024_8.6G	105.53	9.5	39.1
87	SemanticFPN	pt_SemanticFPN_cityscapes_256_512_10G	28.66	34.9	159.3
88	ResNet20-face	pt_facerec-resnet20_mixed_112_96_3.5G	5.74	174.1	309.4
89	face quality	pt_face-quality_80_60_61.68M	0.47	2144.4	6877.7
90	multi_task_v2	pt_MT-resnet18_mixed_320_512_13.65G	30.67	32.6	91.8
91	face_reid_large	pt_facereid-large_96_96_515M	1.13	881.6	2043
92	face_reid_small	pt_facereid-small_80_80_90M	0.52	1909.7	5661.7
93	person_reid	pt_personreid-res50_market1501_256_128_5.4G	9.69	103.2	201.7
94	person_reid	pt_personreid-res18_market1501_176_80_1.1G	2.69	371.4	689.9
95	pointpillars	pt_pointpillars_kitti_12000_100_10.8G	48.88	20.4	49.5
96	salsanext	pt_salsanext_semantic-kitti_64_2048_0.6_20.4G	179.83	5.6	21.4
97	FPN-R18 (light-weight)	pt_FPN-resnet18_covid19-seg_352_352_22.7G	26.35	37.9	79.8
98	2d-unet	pt_unet_chaos-CT_512_512_23.3G	43.61	22.9	59.2
99	surround-view pointpillars	pt_pointpillars_nuscenes_40000_64_108G	429.69	2.3	9.3
100	salsanext_v2	pt_salsanextv2_semantic-kitti_64_2048_0.75_32G	238.18	4.2	11.8
101	centerpoint	pt_centerpoint_astyx_2560_40_54G	59.45	16.8	20.7
102	pointpainting	pt_pointpainting_nuscenes_126G	763.33	1.3	4.7
103	multi_task_v3	pt_multitaskv3_mixed_320_512_25.44G	57.69	17.3	54.4
104	FADnet	pt_fadnet_sceneflow_576_960_359G	-	-	-
105	SA-gate	pt_sa-gate_NYUv2_360_360_178G	-	-	-
106	Bayesian Crowd Counting	pt_BCC_shanghaitech_800_1000_268.9G	287.24	3.5	7.9
107	PMG	pt_pmg_rp2k_224_224_2.28G	6.5	153.7	313.9
108	SemanticFPN-mobilenetv2	pt_SemanticFPN-mobilenetv2_cityscapes_512_1024_5.4G	96.67	10.3	53.5
-	Inception_v3	torchvision_inception_v3	16.01	62.4	122.2
-	SqueezeNet	torchvision_squeezenet	4.37	228.5	744.7
-	resnet50	torchvision_resnet50	13.64	73.3	144

Performance on VCK190

Measured with Vitis AI 1.4 and Vitis AI Library 1.4

Click here to view details

The following table lists the performance number including end-to-end throughput and latency for each model on the Versal board with 96 AIEs running at 1333Mhz:

No.	Model	Name	E2E throughput (fps) Single Thread	E2E throughput (fps) Multi Thread
1	resnet50	cf_resnet50_imagenet_224_224_7.7G	853.5	1381.7
2	resnet18	cf_resnet18_imagenet_224_224_3.65G	1286.7	2955.8
3	Inception_v1	cf_inceptionv1_imagenet_224_224_3.16G	983.3	1750.7
4	Inception_v2	cf_inceptionv2_imagenet_224_224_4G	785.9	1245.6
5	Inception_v3	cf_inceptionv3_imagenet_299_299_11.4G	424.1	611.4
6	Inception_v4	cf_inceptionv4_imagenet_299_299_24.5G	243.1	295.4
7	Mobilenet_v2	cf_mobilenetv2_imagenet_224_224_0.59G	1255.1	2857.2
8	SqueezeNet	cf_squeezenet_imagenet_227_227_0.76G	2431.6	4173.3
9	ssd_pedestrian_pruned_0_97	cf_ssdpedestrian_coco_360_640_0.97_5.9G	317.2	702.8
10	refinedet_baseline	cf_refinedet_coco_360_480_123G	119.6	149.9
11	refinedet_pruned_0_8	cf_refinedet_coco_360_480_0.8_25G	267.1	465.6
12	refinedet_pruned_0_92	cf_refinedet_coco_360_480_0.92_10.10G	339.7	699.1
13	refinedet_pruned_0_96	cf_refinedet_coco_360_480_0.96_5.08G	405.6	853.8
14	ssd_adas_pruned_0_95	cf_ssdadas_bdd_360_480_0.95_6.3G	383	816.1
15	ssd_traffic_pruned_0_9	cf_ssdtraffic_360_480_0.9_11.6G	282.9	665.2
16	VPGnet_pruned_0_99	cf_VPGnet_caltechlane_480_640_0.99_2.5G	269.9	678.8
17	ssd_mobilenet_v2	cf_ssdmobilenetv2_bdd_360_480_6.57G	74.9	167.8
18	FPN	cf_fpn_cityscapes_256_512_8.9G	82.9	190.0
19	SP_net	cf_SPnet_aichallenger_224_128_0.54G	1813.2	4022.4
20	Openpose_pruned_0_3	cf_openpose_aichallenger_368_368_0.3_189.7G	22.1	42.7
21	densebox_320_320	cf_densebox_wider_320_320_0.49G	1018.4	2161
22	densebox_640_360	cf_densebox_wider_360_640_1.11G	499.4	1030.8
23	face_landmark	cf_landmark_celeba_96_72_0.14G	5249.7	10082
24	reid	cf_reid_market1501_160_80_0.95G	2685.6	5065.2
25	multi_task	cf_multitask_bdd_288_512_14.8G	147.3	302.3
26	yolov3_bdd	dk_yolov3_bdd_288_512_53.7G	150.2	200.2
27	yolov3_adas_pruned_0_9	dk_yolov3_cityscapes_256_512_0.9_5.46G	406.7	967.4
28	yolov3_voc	dk_yolov3_voc_416_416_65.42G	156	200.9
29	yolov2_voc	dk_yolov2_voc_448_448_34G	279	460.3
30	yolov2_voc_pruned_0_66	dk_yolov2_voc_448_448_0.66_11.56G	386.8	846.6
31	yolov2_voc_pruned_0_71	dk_yolov2_voc_448_448_0.71_9.86G	406.6	943.8
32	yolov2_voc_pruned_0_77	dk_yolov2_voc_448_448_0.77_7.82G	425.1	1048.6
33	ResNet20-face	cf_facerec-resnet20_112_96_3.5G	1970.9	2555.1
34	ResNet64-face	cf_facerec-resnet64_112_96_11G	1086.7	1244.7
35	FPN_Res18_segmentation	cf_FPN-resnet18_EDD_320_320_45.3G	66.7	188.8
36	plate detection	cf_plate-detection_320_320_0.49G	1199.3	2500.2
37	plate recognition	cf_plate-recognition_96_288_1.75G	457.3	916.1
38	retinaface	cf_retinaface_wider_360_640_1.11G	335.1	750.5
39	face_quality	cf_face-quality_80_60_61.68M	7858.9	18874.4
40	FPN-R18(light-weight)	cf_FPN-resnet18_Endov_240_320_13.75G	102.6	231.9
41	Hourglass	cf_hourglass_mpii_256_256_10.2G	72.8	132
42	tiny-yolov3	dk_tiny-yolov3_416_416_5.46G	521.9	1281.2
43	yolov4	dk_yolov4_coco_416_416_60.1G	118.4	167.8
44	pruned_yolov4	dk_yolov4_coco_416_416_0.36_38.2G	129.8	190.2
45	Inception_resnet_v2	tf_inceptionresnetv2_imagenet_299_299_26.35G	251.1	307
46	Inception_v1	tf_inceptionv1_imagenet_224_224_3G	988.4	1774.6
47	Inception_v3	tf_inceptionv3_imagenet_299_299_11.45G	423.5	610.7
48	Inception_v4	tf_inceptionv4_imagenet_299_299_24.55G	243.4	295.3
49	Mobilenet_v1	tf_mobilenetv1_0.25_imagenet_128_128_27.15M	3614.5	7974.1
50	Mobilenet_v1	tf_mobilenetv1_0.5_imagenet_160_160_150.07M	2594.2	6211.9
51	Mobilenet_v1	tf_mobilenetv1_1.0_imagenet_224_224_1.14G	1383.4	3577.1
52	Mobilenet_v2	tf_mobilenetv2_1.0_imagenet_224_224_0.59G	1253.5	2825.8
53	Mobilenet_v2	tf_mobilenetv2_1.4_imagenet_224_224_1.16G	1045.4	1974.9
54	resnet_v1_50	tf_resnetv1_50_imagenet_224_224_6.97G	904.9	1519.6
55	resnet_v1_101	tf_resnetv1_101_imagenet_224_224_14.4G	607.3	834.5
56	resnet_v1_152	tf_resnetv1_152_imagenet_224_224_21.83G	456.4	574.8
57	vgg_16	tf_vgg16_imagenet_224_224_30.96G	318.6	371.7
58	vgg_19	tf_vgg19_imagenet_224_224_39.28G	286.1	328.2
59	ssd_mobilenet_v1	tf_ssdmobilenetv1_coco_300_300_2.47G	362.8	515.9
60	ssd_mobilenet_v2	tf_ssdmobilenetv2_coco_300_300_3.75G	316.9	508.3
61	ssd_resnet_50_v1_fpn	tf_ssdresnet50v1_fpn_coco_640_640_178.4G	10.1	11.8
62	yolov3_voc	tf_yolov3_voc_416_416_65.63G	156.8	201.5
63	mlperf_ssd_resnet34	tf_mlperf_resnet34_coco_1200_1200_433G	10.6	23.1
64	Inception_v2	tf_inceptionv2_imagenet_224_224_3.88G	559.1	746.8
65	resnet_v2_50	tf_resnetv2_50_imagenet_299_299_13.1G	400.9	563.6
66	resnet_v2_101	tf_resnetv2_101_imagenet_299_299_26.78G	276.5	344.8
67	resnet_v2_152	tf_resnetv2_152_imagenet_299_299_40.47G	210.5	247.9
68	ssdlite_mobilenetv2	tf_ssdlite_mobilenetv2_coco_300_300_1.5G	333.1	507.9
69	ssd_inceptionv2	tf_ssdinceptionv2_coco_300_300_9.62G	157.6	386.4
70	Mobilenet_v2	tf_mobilenetv2_cityscapes_1024_2048_132.74G	4.5	11.6
71	efficientnet-edgetpu-S	tf_efficientnet-edgetpu-S_imagenet_224_224_4.72G	697.3	1015
72	efficientnet-edgetpu-M	tf_efficientnet-edgetpu-M_imagenet_240_240_7.34G	519.5	700.4
73	efficientnet-edgetpu-L	tf_efficientnet-edgetpu-L_imagenet_300_300_19.36G	232.4	279.6
74	mlperf_resnet50	tf_mlperf_resnet50_imagenet_224_224_8.19G	839.9	1345.7
75	refinedet	tf_refinedet_VOC_320_320_81.9G	77.5	186.6
76	mobilenet_edge_1.0	tf_mobilenetEdge1.0_imagenet_224_224_990M	1225.1	2721.8
77	mobilenet_edge_0.75	tf_mobilenetEdge0.75_imagenet_224_224_624M	1302	3117.1
78	refinedet_medical	tf_RefineDet-Medical_EDD_320_320_9.83G	416.7	920
79	pruned_rcan	tf_rcan_DIV2K_360_640_0.98_86.95G	47.5	61.6
80	resnet50	tf2_resnet50_imagenet_224_224_7.76G	854	1379.9
81	Mobilenet_v1	tf2_mobilenetv1_imagenet_224_224_1.15G	1382.6	3578.7
82	Inception_v3	tf2_inceptionv3_imagenet_299_299_11.5G	440.9	647.4
83	2d-unet	tf2_2d-unet_nuclei_128_128_5.31G	1161.8	2033.1
84	ERFNet	tf2_erfnet_cityscapes_512_1024_54G	17.6	45.2
85	efficientnet-b0	tf2_efficientnet-b0_imagenet_224_224_0.36G	687.5	995.4
86	ENet	pt_ENet_cityscapes_512_1024_8.6G	21.1	48
87	SemanticFPN	pt_SemanticFPN_cityscapes_256_512_10G	88.6	191
88	ResNet20-face	pt_facerec-resnet20_mixed_112_96_3.5G	1967.1	2554.8
89	face quality	pt_face-quality_80_60_61.68M	7780	18640.7
90	multi_task_v2	pt_MT-resnet18_mixed_320_512_13.65G	120.8	265.1
91	face_reid_large	pt_facereid-large_96_96_515M	5040.1	12355.3
92	face_reid_small	pt_facereid-small_80_80_90M	7286.5	16493.1
93	person_reid	pt_personreid-res50_market1501_256_128_5.4G	1133.2	1825.2
94	person_reid	pt_personreid-res18_market1501_176_80_1.1G	2750.4	4891.6
95	pointpillars	pt_pointpillars_kitti_12000_100_10.8G	34.7	53.9
96	salsanext	pt_salsanext_semantic-kitti_64_2048_0.6_20.4G	11	22.3
97	FPN-R18 (light-weight)	pt_FPN-resnet18_covid19-seg_352_352_22.7G	366.1	577
98	2d-unet	pt_unet_chaos-CT_512_512_23.3G	83.3	220
99	surround-view pointpillars	pt_pointpillars_nuscenes_40000_64_108G	7.4	17.2
100	salsanext_v2	pt_salsanextv2_semantic-kitti_64_2048_0.75_32G	9.4	21.5
101	centerpoint	pt_centerpoint_astyx_2560_40_54G	111.2	225.4
102	pointpainting	pt_pointpainting_nuscenes_126G	3.5	6.5
103	multi_task_v3	pt_multitaskv3_mixed_320_512_25.44G	71.3	183.9
104	FADnet	pt_fadnet_sceneflow_576_960_359G	5.79	7.19
105	SA-gate	pt_sa-gate_NYUv2_360_360_178G	8.2	14.1
106	Bayesian Crowd Counting	pt_BCC_shanghaitech_800_1000_268.9G	30.6	55.6
107	PMG	pt_pmg_rp2k_224_224_2.28G	1174.1	2057.9
108	SemanticFPN-mobilenetv2	pt_SemanticFPN-mobilenetv2_cityscapes_512_1024_5.4G	21.8	48.2
-	Inception_v3	torchvision_inception_v3	424.1	611.3
-	SqueezeNet	torchvision_squeezenet	807.4	1971.8
-	resnet50	torchvision_resnet50	836.9	1344.6

Performance on Kria KV260 SOM

Measured with Vitis AI 1.4 and Vitis AI Library 1.4

Click here to view details

The following table lists the performance number including end-to-end throughput and latency for each model on the Kria KV260 board with a 1 * B4096F @ 300MHz DPU configuration:

No.	Model	Name	E2E latency (ms) Thread Num =1	E2E throughput (fps) Single Thread	E2E throughput (fps) Multi Thread
1	resnet50	cf_resnet50_imagenet_224_224_7.7G	13.68	73	77.1
2	resnet18	cf_resnet18_imagenet_224_224_3.65G	5.37	186.1	213.8
3	Inception_v1	cf_inceptionv1_imagenet_224_224_3.16G	5.46	183.2	210.6
4	Inception_v2	cf_inceptionv2_imagenet_224_224_4G	7.53	132.8	147.8
5	Inception_v3	cf_inceptionv3_imagenet_299_299_11.4G	16.86	59.3	63.7
6	Inception_v4	cf_inceptionv4_imagenet_299_299_24.5G	34.32	29.1	30.1
7	Mobilenet_v2	cf_mobilenetv2_imagenet_224_224_0.59G	3.95	252.9	310.4
8	SqueezeNet	cf_squeezenet_imagenet_227_227_0.76G	3.7	269.9	557.7
9	ssd_pedestrian_pruned_0_97	cf_ssdpedestrian_coco_360_640_0.97_5.9G	12.76	78.3	107
10	refinedet_baseline	cf_refinedet_coco_360_480_123G	111.02	9	9.2
11	refinedet_pruned_0_8	cf_refinedet_coco_360_480_0.8_25G	29.44	34	37
12	refinedet_pruned_0_92	cf_refinedet_coco_360_480_0.92_10.10G	15.58	64.2	75.9
13	refinedet_pruned_0_96	cf_refinedet_coco_360_480_0.96_5.08G	11.3	88.5	111.2
14	ssd_adas_pruned_0_95	cf_ssdadas_bdd_360_480_0.95_6.3G	11.18	89.4	119.2
15	ssd_traffic_pruned_0_9	cf_ssdtraffic_360_480_0.9_11.6G	17.65	56.6	74.3
16	VPGnet_pruned_0_99	cf_VPGnet_caltechlane_480_640_0.99_2.5G	10.6	94.4	149.4
17	ssd_mobilenet_v2	cf_ssdmobilenetv2_bdd_360_480_6.57G	40.41	24.7	61.7
18	FPN	cf_fpn_cityscapes_256_512_8.9G	29.41	34	76.4
19	SP_net	cf_SPnet_aichallenger_224_128_0.54G	1.88	530.4	680.3
20	Openpose_pruned_0_3	cf_openpose_aichallenger_368_368_0.3_189.7G	274.99	3.6	5.5
21	densebox_320_320	cf_densebox_wider_320_320_0.49G	2.37	421.7	845.4
22	densebox_640_360	cf_densebox_wider_360_640_1.11G	4.85	206	418.5
23	face_landmark	cf_landmark_celeba_96_72_0.14G	1.25	801.4	938.1
24	reid	cf_reid_market1501_160_80_0.95G	3.02	331.2	379.1
25	multi_task	cf_multitask_bdd_288_512_14.8G	26.07	38.3	53.7
26	yolov3_bdd	dk_yolov3_bdd_288_512_53.7G	77.06	13	13.5
27	yolov3_adas_pruned_0_9	dk_yolov3_cityscapes_256_512_0.9_5.46G	11.03	90.6	122.8
28	yolov3_voc	dk_yolov3_voc_416_416_65.42G	74.85	13.3	13.8
29	yolov2_voc	dk_yolov2_voc_448_448_34G	36.91	27.1	29
30	yolov2_voc_pruned_0_66	dk_yolov2_voc_448_448_0.66_11.56G	15.15	66	76.7
31	yolov2_voc_pruned_0_71	dk_yolov2_voc_448_448_0.71_9.86G	13.2	75.7	90.4
32	yolov2_voc_pruned_0_77	dk_yolov2_voc_448_448_0.77_7.82G	11.36	88	108.5
33	ResNet20-face	cf_facerec-resnet20_112_96_3.5G	6.21	161.1	167
34	ResNet64-face	cf_facerec-resnet64_112_96_11G	13.71	72.9	74.1
35	FPN_Res18_segmentation	cf_FPN-resnet18_EDD_320_320_45.3G	78.11	12.8	16.8
36	plate detection	cf_plate-detection_320_320_0.49G	1.96	510.8	1059.4
37	plate recognition	cf_plate-recognition_96_288_1.75G	6.06	164.9	263.9
38	retinaface	cf_retinaface_wider_360_640_1.11G	7.85	127.4	267
39	face_quality	cf_face-quality_80_60_61.68M	0.45	2231	3736.3
40	FPN-R18(light-weight)	cf_FPN-resnet18_Endov_240_320_13.75G	28.8	34.7	64.1
41	Hourglass	cf_hourglass_mpii_256_256_10.2G	55.73	17.9	52.2
42	tiny-yolov3	dk_tiny-yolov3_416_416_5.46G	8.11	123.3	165.7
43	yolov4	dk_yolov4_coco_416_416_60.1G	72.21	13.8	14.8
44	pruned_yolov4	dk_yolov4_coco_416_416_0.36_38.2G	53.7	18.6	20.5
45	Inception_resnet_v2	tf_inceptionresnetv2_imagenet_299_299_26.35G	44.17	22.6	23.2
46	Inception_v1	tf_inceptionv1_imagenet_224_224_3G	5.38	185.7	214.7
47	Inception_v3	tf_inceptionv3_imagenet_299_299_11.45G	16.91	59.1	63.5
48	Inception_v4	tf_inceptionv4_imagenet_299_299_24.55G	34.35	29.1	30.1
49	Mobilenet_v1	tf_mobilenetv1_0.25_imagenet_128_128_27.15M	0.88	1133.2	2053.8
50	Mobilenet_v1	tf_mobilenetv1_0.5_imagenet_160_160_150.07M	1.38	725.8	1127.5
51	Mobilenet_v1	tf_mobilenetv1_1.0_imagenet_224_224_1.14G	3.3	302.9	387.5
52	Mobilenet_v2	tf_mobilenetv2_1.0_imagenet_224_224_0.59G	4.05	247	299.9
53	Mobilenet_v2	tf_mobilenetv2_1.4_imagenet_224_224_1.16G	5.52	181.1	208.2
54	resnet_v1_50	tf_resnetv1_50_imagenet_224_224_6.97G	12.61	79.3	84.1
55	resnet_v1_101	tf_resnetv1_101_imagenet_224_224_14.4G	23.13	43.2	44.6
56	resnet_v1_152	tf_resnetv1_152_imagenet_224_224_21.83G	33.56	29.8	30.4
57	vgg_16	tf_vgg16_imagenet_224_224_30.96G	52.04	19.2	19.5
58	vgg_19	tf_vgg19_imagenet_224_224_39.28G	59.56	16.8	17
59	ssd_mobilenet_v1	tf_ssdmobilenetv1_coco_300_300_2.47G	9.09	110	163.2
60	ssd_mobilenet_v2	tf_ssdmobilenetv2_coco_300_300_3.75G	12.55	79.7	102.3
61	ssd_resnet_50_v1_fpn	tf_ssdresnet50v1_fpn_coco_640_640_178.4G	361.24	2.8	5.2
62	yolov3_voc	tf_yolov3_voc_416_416_65.63G	74.85	13.3	13.8
63	mlperf_ssd_resnet34	tf_mlperf_resnet34_coco_1200_1200_433G	555.45	1.8	2.6
64	Inception_v2	tf_inceptionv2_imagenet_224_224_3.88G	10.66	93.8	100.6
65	resnet_v2_50	tf_resnetv2_50_imagenet_299_299_13.1G	27.89	35.8	42.8
66	resnet_v2_101	tf_resnetv2_101_imagenet_299_299_26.78G	48.29	20.7	23.6
67	resnet_v2_152	tf_resnetv2_152_imagenet_299_299_40.47G	67.53	14.8	16.1
68	ssdlite_mobilenetv2	tf_ssdlite_mobilenetv2_coco_300_300_1.5G	9.77	102.3	143.1
69	ssd_inceptionv2	tf_ssdinceptionv2_coco_300_300_9.62G	25.4	39.3	44.9
70	Mobilenet_v2	tf_mobilenetv2_cityscapes_1024_2048_132.74G	621.96	1.6	2.7
71	efficientnet-edgetpu-S	tf_efficientnet-edgetpu-S_imagenet_224_224_4.72G	10.32	96.9	104.4
72	efficientnet-edgetpu-M	tf_efficientnet-edgetpu-M_imagenet_240_240_7.34G	14.22	70.3	74.4
73	efficientnet-edgetpu-L	tf_efficientnet-edgetpu-L_imagenet_300_300_19.36G	35.19	28.4	31.2
74	mlperf_resnet50	tf_mlperf_resnet50_imagenet_224_224_8.19G	13.92	71.8	75.8
75	refinedet	tf_refinedet_VOC_320_320_81.9G	94.85	10.5	13
76	mobilenet_edge_1.0	tf_mobilenetEdge1.0_imagenet_224_224_990M	5.00	199.7	234
77	mobilenet_edge_0.75	tf_mobilenetEdge0.75_imagenet_224_224_624M	4.14	241.3	291.9
78	refinedet_medical	tf_RefineDet-Medical_EDD_320_320_9.83G	14.5	69	84.3
79	pruned_rcan	tf_rcan_DIV2K_360_640_0.98_86.95G	132.69	7.5	7.8
80	resnet50	tf2_resnet50_imagenet_224_224_7.76G	14.11	70.8	74.9
81	Mobilenet_v1	tf2_mobilenetv1_imagenet_224_224_1.15G	3.35	298.6	380.6
82	Inception_v3	tf2_inceptionv3_imagenet_299_299_11.5G	17.03	58.7	63
83	2d-unet	tf2_2d-unet_nuclei_128_128_5.31G	6.9	144.8	158.3
84	ERFNet	tf2_erfnet_cityscapes_512_1024_54G	145.22	6.9	12.8
85	efficientnet-b0	tf2_efficientnet-b0_imagenet_224_224_0.36G	-	-	-
86	ENet	pt_ENet_cityscapes_512_1024_8.6G	111.6	9	21
87	SemanticFPN	pt_SemanticFPN_cityscapes_256_512_10G	29.41	34	76.6
88	ResNet20-face	pt_facerec-resnet20_mixed_112_96_3.5G	6.21	161	167.1
89	face quality	pt_face-quality_80_60_61.68M	0.49	2047.3	3742.3
90	multi_task_v2	pt_MT-resnet18_mixed_320_512_13.65G	33.1	30.2	42.8
91	face_reid_large	pt_facereid-large_96_96_515M	1.19	839.6	1058.4
92	face_reid_small	pt_facereid-small_80_80_90M	0.53	1893.6	3055.3
93	person_reid	pt_personreid-res50_market1501_256_128_5.4G	10.42	96	102.3
94	person_reid	pt_personreid-res18_market1501_176_80_1.1G	3.03	329.9	368.9
95	pointpillars	pt_pointpillars_kitti_12000_100_10.8G	48.95	20.4	28.8
96	salsanext	pt_salsanext_semantic-kitti_64_2048_0.6_20.4G	181.06	5.5	18.6
97	FPN-R18 (light-weight)	pt_FPN-resnet18_covid19-seg_352_352_22.7G	26.63	37.5	39.9
98	2d-unet	pt_unet_chaos-CT_512_512_23.3G	54.58	18.3	22.8
99	surround-view pointpillars	pt_pointpillars_nuscenes_40000_64_108G	465.21	2.1	5
100	salsanext_v2	pt_salsanextv2_semantic-kitti_64_2048_0.75_32G	246.92	4.0	10.1
101	centerpoint	pt_centerpoint_astyx_2560_40_54G	784.6	1.3	4.2
102	pointpainting	pt_pointpainting_nuscenes_126G	820.21	1.2	2.6
103	multi_task_v3	pt_multitaskv3_mixed_320_512_25.44G	61.39	16.3	25.9
104	FADnet	pt_fadnet_sceneflow_576_960_359G	-	-	-
105	SA-gate	pt_sa-gate_NYUv2_360_360_178G	-	-	-
106	Bayesian Crowd Counting	pt_BCC_shanghaitech_800_1000_268.9G	292.42	3.4	3.9
107	PMG	pt_pmg_rp2k_224_224_2.28G	6.82	146.5	160.1
108	SemanticFPN-mobilenetv2	pt_SemanticFPN-mobilenetv2_cityscapes_512_1024_5.4G	100.17	10	28.1
-	Inception_v3	torchvision_inception_v3	16.85	59.3	63.8
-	SqueezeNet	torchvision_squeezenet	4.52	221.2	385
-	resnet50	torchvision_resnet50	14.47	69.1	72.7

Performance on VCK5000

Measured with Vitis AI 1.4 and Vitis AI Library 1.4

Click here to view details

The following table lists the performance number including end-to-end throughput for models on the Versal ACAP VCK5000 board with DPUCVDX8H running at 8PE@350 MHz in Gen3x16:

No.	Model	Name	DPU Frequency(MHz)	E2E throughput (fps) Multi Thread
1	resnet50	cf_resnet50_imagenet_224_224_7.7G	350	4522.4
2	resnet18	cf_resnet18_imagenet_224_224_3.65G	350	6442.1
3	Inception_v1	cf_inceptionv1_imagenet_224_224_3.16G	350	4099.3
4	SqueezeNet	cf_squeezenet_imagenet_227_227_0.76G	350	8256.3
5	refinedet_baseline	cf_refinedet_coco_360_480_123G	350	291.8
6	refinedet_pruned_0_8	cf_refinedet_coco_360_480_0.8_25G	350	654.3
7	refinedet_pruned_0_92	cf_refinedet_coco_360_480_0.92_10.10G	350	845.8
8	refinedet_pruned_0_96	cf_refinedet_coco_360_480_0.96_5.08G	350	888.4
9	ssd_adas_pruned_0_95	cf_ssdadas_bdd_360_480_0.95_6.3G	350	930.6
10	ssd_traffic_pruned_0_9	cf_ssdtraffic_360_480_0.9_11.6G	350	894.5
11	VPGnet_pruned_0_99	cf_VPGnet_caltechlane_480_640_0.99_2.5G	350	600.8
12	FPN	cf_fpn_cityscapes_256_512_8.9G	350	1107.7
13	SP_net	cf_SPnet_aichallenger_224_128_0.54G	350	7649.9
14	Openpose_pruned_0_3	cf_openpose_aichallenger_368_368_0.3_189.7G	350	168.6
15	densebox_320_320	cf_densebox_wider_320_320_0.49G	350	5912.0
16	densebox_640_360	cf_densebox_wider_360_640_1.11G	350	2814.7
17	face_landmark	cf_landmark_celeba_96_72_0.14G	350	14238.6
18	reid	cf_reid_market1501_160_80_0.95G	350	10498.1
19	multi_task	cf_multitask_bdd_288_512_14.8G	350	699.0
20	yolov3_bdd	dk_yolov3_bdd_288_512_53.7G	350	383.0
21	yolov3_adas_pruned_0_9	dk_yolov3_cityscapes_256_512_0.9_5.46G	350	1379.1
22	yolov3_voc	dk_yolov3_voc_416_416_65.42G	350	458.6
23	yolov2_voc	dk_yolov2_voc_448_448_34G	350	950.7
24	yolov2_voc_pruned_0_66	dk_yolov2_voc_448_448_0.66_11.56G	350	1572.7
25	yolov2_voc_pruned_0_71	dk_yolov2_voc_448_448_0.71_9.86G	350	1726.9
26	yolov2_voc_pruned_0_77	dk_yolov2_voc_448_448_0.77_7.82G	350	1701.5
27	FPN_Res18_segmentation	cf_FPN-resnet18_EDD_320_320_45.3G	350	536.2
28	plate detection	cf_plate-detection_320_320_0.49G	350	7793.1
29	tiny-yolov3	dk_tiny-yolov3_416_416_5.46G	350	2605.0
30	Inception_v1	tf_inceptionv1_imagenet_224_224_3G	350	4358.7
31	resnet_v1_50	tf_resnetv1_50_imagenet_224_224_6.97G	350	4852.5
32	resnet_v1_101	tf_resnetv1_101_imagenet_224_224_14.4G	350	2937.2
33	resnet_v1_152	tf_resnetv1_152_imagenet_224_224_21.83G	350	2096.9
34	yolov3_voc	tf_yolov3_voc_416_416_65.63G	350	459.5
35	mlperf_resnet50	tf_mlperf_resnet50_imagenet_224_224_8.19G	350	4431.5
36	refinedet	tf_refinedet_VOC_320_320_81.9G	350	378.8
37	refinedet_medical	tf_RefineDet-Medical_EDD_320_320_9.83G	350	1289.3
38	resnet50	tf2_resnet50_imagenet_224_224_7.76G	350	4461.0
39	2d-unet	tf2_2d-unet_nuclei_128_128_5.31G	350	1962.7
40	ENet	pt_ENet_cityscapes_512_1024_8.6G	350	140.8
41	SemanticFPN	pt_SemanticFPN_cityscapes_256_512_10G	350	1120.0
42	face quality	pt_face-quality_80_60_61.68M	350	41906.7
43	salsanext	pt_salsanext_semantic-kitti_64_2048_0.6_20.4G	350	177.5
44	FPN-R18 (light-weight)	pt_FPN-resnet18_covid19-seg_352_352_22.7G	350	1079.5
45	2d-unet	pt_unet_chaos-CT_512_512_23.3G	350	228.8
-	resnet50	torchvision_resnet50	350	4517.3
-	SqueezeNet	torchvision_squeezenet	350	5147.7

Performance on U50

Measured with Vitis AI 1.4 and Vitis AI Library 1.4

Click here to view details

The following table lists the performance number including end-to-end throughput for each model on the Alveo U50 board with 6 DPUCAHX8H kernels running at 300Mhz in Gen3x4:

No.	Model	Name	DPU Frequency(MHz)	E2E throughput (fps) Multi Thread
1	resnet50	cf_resnet50_imagenet_224_224_7.7G	300	573.2
2	resnet18	cf_resnet18_imagenet_224_224_3.65G	300	1410.1
3	Inception_v1	cf_inceptionv1_imagenet_224_224_3.16G	300	1234.0
4	Inception_v2	cf_inceptionv2_imagenet_224_224_4G	300	973.4
5	Inception_v3	cf_inceptionv3_imagenet_299_299_11.4G	300	406.6
6	Inception_v4	cf_inceptionv4_imagenet_299_299_24.5G	300	185.6
7	SqueezeNet	cf_squeezenet_imagenet_227_227_0.76G	300	3619.4
8	ssd_pedestrian_pruned_0_97	cf_ssdpedestrian_coco_360_640_0.97_5.9G	270	568.7
9	refinedet_baseline	cf_refinedet_coco_360_480_123G	270	49.8
10	refinedet_pruned_0_8	cf_refinedet_coco_360_480_0.8_25G	270	193.1
11	refinedet_pruned_0_92	cf_refinedet_coco_360_480_0.92_10.10G	270	404.9
12	refinedet_pruned_0_96	cf_refinedet_coco_360_480_0.96_5.08G	270	592.5
13	ssd_adas_pruned_0_95	cf_ssdadas_bdd_360_480_0.95_6.3G	300	656.0
14	ssd_traffic_pruned_0_9	cf_ssdtraffic_360_480_0.9_11.6G	300	428.9
15	VPGnet_pruned_0_99	cf_VPGnet_caltechlane_480_640_0.99_2.5G	270	481.0
16	FPN	cf_fpn_cityscapes_256_512_8.9G	300	440.3
17	SP_net	cf_SPnet_aichallenger_224_128_0.54G	300	2948.1
18	Openpose_pruned_0_3	cf_openpose_aichallenger_368_368_0.3_189.7G	270	29.0
19	densebox_320_320	cf_densebox_wider_320_320_0.49G	300	2232.1
20	densebox_640_360	cf_densebox_wider_360_640_1.11G	300	971.6
21	face_landmark	cf_landmark_celeba_96_72_0.14G	300	10098.8
22	reid	cf_reid_market1501_160_80_0.95G	300	3849.7
23	multi_task	cf_multitask_bdd_288_512_14.8G	300	336.4
24	yolov3_bdd	dk_yolov3_bdd_288_512_53.7G	270	74.8
25	yolov3_adas_pruned_0_9	dk_yolov3_cityscapes_256_512_0.9_5.46G	270	611.5
26	yolov3_voc	dk_yolov3_voc_416_416_65.42G	270	77.0
27	yolov2_voc	dk_yolov2_voc_448_448_34G	270	162.5
28	yolov2_voc_pruned_0_66	dk_yolov2_voc_448_448_0.66_11.56G	270	410.9
29	yolov2_voc_pruned_0_71	dk_yolov2_voc_448_448_0.71_9.86G	270	481.5
30	yolov2_voc_pruned_0_77	dk_yolov2_voc_448_448_0.77_7.82G	270	585.5
31	ResNet20-face	cf_facerec-resnet20_112_96_3.5G	300	1331.0
32	ResNet64-face	cf_facerec-resnet64_112_96_11G	300	491.6
33	FPN_Res18_segmentation	cf_FPN-resnet18_EDD_320_320_45.3G	300	101.5
34	plate detection	cf_plate-detection_320_320_0.49G	300	6083.8
35	plate recognition	cf_plate-recognition_96_288_1.75G	270	1215.5
36	face_quality	cf_face-quality_80_60_61.68M	270	22096.9
37	tiny-yolov3	dk_tiny-yolov3_416_416_5.46G	270	870.7
38	yolov4	dk_yolov4_coco_416_416_60.1G	270	81.0
39	pruned_yolov4	dk_yolov4_coco_416_416_0.36_38.2G	300	84.9
40	Inception_resnet_v2	tf_inceptionresnetv2_imagenet_299_299_26.35G	300	172.2
41	Inception_v1	tf_inceptionv1_imagenet_224_224_3G	300	1252.6
42	Inception_v3	tf_inceptionv3_imagenet_299_299_11.45G	300	407.4
43	Inception_v4	tf_inceptionv4_imagenet_299_299_24.55G	300	185.6
44	resnet_v1_50	tf_resnetv1_50_imagenet_224_224_6.97G	300	642.9
45	resnet_v1_101	tf_resnetv1_101_imagenet_224_224_14.4G	300	333.8
46	resnet_v1_152	tf_resnetv1_152_imagenet_224_224_21.83G	300	222.7
47	vgg_16	tf_vgg16_imagenet_224_224_30.96G	300	161.6
48	vgg_19	tf_vgg19_imagenet_224_224_39.28G	300	134.6
49	ssd_resnet_50_v1_fpn	tf_ssdresnet50v1_fpn_coco_640_640_178.4G	270	31.2
50	yolov3_voc	tf_yolov3_voc_416_416_65.63G	270	77.1
51	mlperf_ssd_resnet34	tf_mlperf_resnet34_coco_1200_1200_433G	270	13.8
52	mlperf_resnet50	tf_mlperf_resnet50_imagenet_224_224_8.19G	270	500.3
53	refinedet	tf_refinedet_VOC_320_320_81.9G	270	71.3
54	refinedet_medical	tf_RefineDet-Medical_EDD_320_320_9.83G	270	421.1
55	pruned_rcan	tf_rcan_DIV2K_360_640_0.98_86.95G	300	49.5
56	resnet50	tf2_resnet50_imagenet_224_224_7.76G	270	516.6
57	Inception_v3	tf2_inceptionv3_imagenet_299_299_11.5G	270	372.4
58	2d-unet	tf2_2d-unet_nuclei_128_128_5.31G	270	1051.5
59	ERFNet	tf2_erfnet_cityscapes_512_1024_54G	270	49.6
60	ENet	pt_ENet_cityscapes_512_1024_8.6G	270	71.3
61	SemanticFPN	pt_SemanticFPN_cityscapes_256_512_10G	270	425.3
62	ResNet20-face	pt_facerec-resnet20_mixed_112_96_3.5G	270	1213.2
63	face quality	pt_face-quality_80_60_61.68M	270	21970.7
64	face_reid_large	pt_facereid-large_96_96_515M	270	7101.2
65	face_reid_small	pt_facereid-small_80_80_90M	270	19689.7
66	person_reid	pt_personreid-res50_market1501_256_128_5.4G	270	805.9
67	person_reid	pt_personreid-res18_market1501_176_80_1.1G	270	3407.2
68	salsanext	pt_salsanext_semantic-kitti_64_2048_0.6_20.4G	270	168.1
69	FPN-R18 (light-weight)	pt_FPN-resnet18_covid19-seg_352_352_22.7G	270	205.0
70	2d-unet	pt_unet_chaos-CT_512_512_23.3G	270	60.2
71	salsanext_v2	pt_salsanextv2_semantic-kitti_64_2048_0.75_32G	300	25.4
72	PMG	pt_pmg_rp2k_224_224_2.28G	300	1088.8
-	Inception_v3	torchvision_inception_v3	300	407.4
-	resnet50	torchvision_resnet50	300	555.1
-	SqueezeNet	torchvision_squeezenet	300	2079.3

The following table lists the performance number including end-to-end throughput for each model on the Alveo U50 board with 1 DPUCAHX8L kernel running at 333Mhz in Gen3x4:

No.	Model	Name	DPU Frequency(MHz)	E2E throughput (fps) Multi Thread
1	resnet50	cf_resnet50_imagenet_224_224_7.7G	333	87.8
2	resnet18	cf_resnet18_imagenet_224_224_3.65G	333	320.9
3	Inception_v1	cf_inceptionv1_imagenet_224_224_3.16G	333	343.7
4	Inception_v2	cf_inceptionv2_imagenet_224_224_4G	333	197.4
5	Inception_v3	cf_inceptionv3_imagenet_299_299_11.4G	333	107.9
6	Inception_v4	cf_inceptionv4_imagenet_299_299_24.5G	333	56.9
7	Mobilenet_v2	cf_mobilenetv2_imagenet_224_224_0.59G	333	1179.5
8	SqueezeNet	cf_squeezenet_imagenet_227_227_0.76G	333	838.0
9	ssd_pedestrian_pruned_0_97	cf_ssdpedestrian_coco_360_640_0.97_5.9G	333	107.7
10	refinedet_baseline	cf_refinedet_coco_360_480_123G	333	30.5
11	refinedet_pruned_0_8	cf_refinedet_coco_360_480_0.8_25G	333	67.3
12	refinedet_pruned_0_92	cf_refinedet_coco_360_480_0.92_10.10G	333	77.4
13	refinedet_pruned_0_96	cf_refinedet_coco_360_480_0.96_5.08G	333	94.6
14	ssd_adas_pruned_0_95	cf_ssdadas_bdd_360_480_0.95_6.3G	333	117.3
15	ssd_traffic_pruned_0_9	cf_ssdtraffic_360_480_0.9_11.6G	333	112.6
16	VPGnet_pruned_0_99	cf_VPGnet_caltechlane_480_640_0.99_2.5G	333	72.3
17	ssd_mobilenet_v2	cf_ssdmobilenetv2_bdd_360_480_6.57G	333	157.7
18	FPN	cf_fpn_cityscapes_256_512_8.9G	333	39.5
19	SP_net	cf_SPnet_aichallenger_224_128_0.54G	333	10966.0
20	Openpose_pruned_0_3	cf_openpose_aichallenger_368_368_0.3_189.7G	333	17.4
21	face_landmark	cf_landmark_celeba_96_72_0.14G	333	5054.6
22	reid	cf_reid_market1501_160_80_0.95G	333	612.5
23	multi_task	cf_multitask_bdd_288_512_14.8G	333	21.4
24	ResNet20-face	cf_facerec-resnet20_112_96_3.5G	333	318.5
25	ResNet64-face	cf_facerec-resnet64_112_96_11G	333	141.1
26	FPN_Res18_segmentation	cf_FPN-resnet18_EDD_320_320_45.3G	333	14.9
27	plate detection	cf_plate-detection_320_320_0.49G	333	1035.2
28	retinaface	cf_retinaface_wider_360_640_1.11G	333	206.4
29	face_quality	cf_face-quality_80_60_61.68M	333	7276.8
30	Inception_resnet_v2	tf_inceptionresnetv2_imagenet_299_299_26.35G	333	35.3
31	Inception_v1	tf_inceptionv1_imagenet_224_224_3G	333	348.7
32	Inception_v3	tf_inceptionv3_imagenet_299_299_11.45G	333	107.7
33	Inception_v4	tf_inceptionv4_imagenet_299_299_24.55G	333	56.7
34	Mobilenet_v1	tf_mobilenetv1_0.5_imagenet_160_160_150.07M	333	5506.2
35	Mobilenet_v1	tf_mobilenetv1_1.0_imagenet_224_224_1.14G	333	2142.3
36	Mobilenet_v2	tf_mobilenetv2_1.0_imagenet_224_224_0.59G	333	1176.7
37	Mobilenet_v2	tf_mobilenetv2_1.4_imagenet_224_224_1.16G	333	867.4
38	resnet_v1_50	tf_resnetv1_50_imagenet_224_224_6.97G	333	102.5
39	resnet_v1_101	tf_resnetv1_101_imagenet_224_224_14.4G	333	56.3
40	resnet_v1_152	tf_resnetv1_152_imagenet_224_224_21.83G	333	38.5
41	vgg_16	tf_vgg16_imagenet_224_224_30.96G	333	57.7
42	vgg_19	tf_vgg19_imagenet_224_224_39.28G	333	52.0
43	ssd_mobilenet_v1	tf_ssdmobilenetv1_coco_300_300_2.47G	333	1051.3
44	ssd_mobilenet_v2	tf_ssdmobilenetv2_coco_300_300_3.75G	333	237.9
45	ssdlite_mobilenetv2	tf_ssdlite_mobilenetv2_coco_300_300_1.5G	333	612.2
46	mlperf_ssd_resnet34	tf_mlperf_resnet34_coco_1200_1200_433G	333	7.0
47	mlperf_resnet50	tf_mlperf_resnet50_imagenet_224_224_8.19G	333	88.5
48	refinedet	tf_refinedet_VOC_320_320_81.9G	333	47.9
49	refinedet_medical	tf_RefineDet-Medical_EDD_320_320_9.83G	333	151.9
50	pruned_rcan	tf_rcan_DIV2K_360_640_0.98_86.95G	333	4.1
51	resnet50	tf2_resnet50_imagenet_224_224_7.76G	333	89.6
52	Mobilenet_v1	tf2_mobilenetv1_imagenet_224_224_1.15G	333	2096.1
53	Inception_v3	tf2_inceptionv3_imagenet_299_299_11.5G	333	102.0
54	2d-unet	tf2_2d-unet_nuclei_128_128_5.31G	333	118.0
55	ERFNet	tf2_erfnet_cityscapes_512_1024_54G	333	5.6
56	ENet	pt_ENet_cityscapes_512_1024_8.6G	333	6.0
57	SemanticFPN	pt_SemanticFPN_cityscapes_256_512_10G	333	36.5
58	ResNet20-face	pt_facerec-resnet20_mixed_112_96_3.5G	333	317.1
59	face quality	pt_face-quality_80_60_61.68M	333	7285.6
60	face_reid_small	pt_facereid-small_80_80_90M	333	4541.7
61	person_reid	pt_personreid-res50_market1501_256_128_5.4G	333	106.4
62	salsanext	pt_salsanext_semantic-kitti_64_2048_0.6_20.4G	333	11.7
63	FPN-R18 (light-weight)	pt_FPN-resnet18_covid19-seg_352_352_22.7G	333	90.3
64	SemanticFPN-mobilenetv2	pt_SemanticFPN-mobilenetv2_cityscapes_512_1024_5.4G	333	11.1
-	Inception_v3	torchvision_inception_v3	333	107.4
-	resnet50	torchvision_resnet50	333	89.1
-	SqueezeNet	torchvision_squeezenet	333	471.2

Performance on U50 lv

Measured with Vitis AI 1.4 and Vitis AI Library 1.4

Click here to view details

The following table lists the performance number including end-to-end throughput for each model on the Alveo U50 board with 10 DPUCAHX8H kernels running at 275Mhz in Gen3x4:

No.	Model	Name	DPU Frequency(MHz)	E2E throughput (fps) Multi Thread
1	resnet50	cf_resnet50_imagenet_224_224_7.7G	247.5	788.3
2	resnet18	cf_resnet18_imagenet_224_224_3.65G	247.5	1938.0
3	Inception_v1	cf_inceptionv1_imagenet_224_224_3.16G	247.5	1702.0
4	Inception_v2	cf_inceptionv2_imagenet_224_224_4G	247.5	1343.2
5	Inception_v3	cf_inceptionv3_imagenet_299_299_11.4G	220	499.1
6	Inception_v4	cf_inceptionv4_imagenet_299_299_24.5G	220	226.8
7	SqueezeNet	cf_squeezenet_imagenet_227_227_0.76G	247.5	4826.3
8	ssd_pedestrian_pruned_0_97	cf_ssdpedestrian_coco_360_640_0.97_5.9G	247.5	863.6
9	refinedet_baseline	cf_refinedet_coco_360_480_123G	192.5	59.2
10	refinedet_pruned_0_8	cf_refinedet_coco_360_480_0.8_25G	247.5	295.8
11	refinedet_pruned_0_92	cf_refinedet_coco_360_480_0.92_10.10G	247.5	616.4
12	refinedet_pruned_0_96	cf_refinedet_coco_360_480_0.96_5.08G	247.5	902.6
13	ssd_adas_pruned_0_95	cf_ssdadas_bdd_360_480_0.95_6.3G	220	820.9
14	ssd_traffic_pruned_0_9	cf_ssdtraffic_360_480_0.9_11.6G	275	653.4
15	VPGnet_pruned_0_99	cf_VPGnet_caltechlane_480_640_0.99_2.5G	247.5	719.8
16	FPN	cf_fpn_cityscapes_256_512_8.9G	247.5	606.1
17	SP_net	cf_SPnet_aichallenger_224_128_0.54G	275	4291.3
18	Openpose_pruned_0_3	cf_openpose_aichallenger_368_368_0.3_189.7G	192.5	34.8
19	densebox_320_320	cf_densebox_wider_320_320_0.49G	275	3335.5
20	densebox_640_360	cf_densebox_wider_360_640_1.11G	275	1449.0
21	face_landmark	cf_landmark_celeba_96_72_0.14G	275	14938.1
22	reid	cf_reid_market1501_160_80_0.95G	275	5833.1
23	multi_task	cf_multitask_bdd_288_512_14.8G	247.5	458.5
24	yolov3_bdd	dk_yolov3_bdd_288_512_53.7G	220	101.6
25	yolov3_adas_pruned_0_9	dk_yolov3_cityscapes_256_512_0.9_5.46G	220	836.5
26	yolov3_voc	dk_yolov3_voc_416_416_65.42G	220	104.7
27	yolov2_voc	dk_yolov2_voc_448_448_34G	220	220.5
28	yolov2_voc_pruned_0_66	dk_yolov2_voc_448_448_0.66_11.56G	220	558.7
29	yolov2_voc_pruned_0_71	dk_yolov2_voc_448_448_0.71_9.86G	220	654.8
30	yolov2_voc_pruned_0_77	dk_yolov2_voc_448_448_0.77_7.82G	220	794.6
31	ResNet20-face	cf_facerec-resnet20_112_96_3.5G	275	2026.2
32	ResNet64-face	cf_facerec-resnet64_112_96_11G	275	749.7
33	FPN_Res18_segmentation	cf_FPN-resnet18_EDD_320_320_45.3G	247.5	140.2
34	plate detection	cf_plate-detection_320_320_0.49G	275	6997.9
35	plate recognition	cf_plate-recognition_96_288_1.75G	247.5	1706.2
36	face_quality	cf_face-quality_80_60_61.68M	247.5	23500.4
37	tiny-yolov3	dk_tiny-yolov3_416_416_5.46G	247.5	1325.7
38	yolov4	dk_yolov4_coco_416_416_60.1G	220	109.9
39	pruned_yolov4	dk_yolov4_coco_416_416_0.36_38.2G	220	111.8
40	Inception_resnet_v2	tf_inceptionresnetv2_imagenet_299_299_26.35G	220	211.0
41	Inception_v1	tf_inceptionv1_imagenet_224_224_3G	247.5	1726.4
42	Inception_v3	tf_inceptionv3_imagenet_299_299_11.45G	220	499.4
43	Inception_v4	tf_inceptionv4_imagenet_299_299_24.55G	220	227.7
44	resnet_v1_50	tf_resnetv1_50_imagenet_224_224_6.97G	247.5	885.3
45	resnet_v1_101	tf_resnetv1_101_imagenet_224_224_14.4G	247.5	459.6
46	resnet_v1_152	tf_resnetv1_152_imagenet_224_224_21.83G	247.5	306.3
47	vgg_16	tf_vgg16_imagenet_224_224_30.96G	247.5	225.4
48	vgg_19	tf_vgg19_imagenet_224_224_39.28G	247.5	187.4
49	ssd_resnet_50_v1_fpn	tf_ssdresnet50v1_fpn_coco_640_640_178.4G	220	42.4
50	yolov3_voc	tf_yolov3_voc_416_416_65.63G	220	104.7
51	mlperf_ssd_resnet34	tf_mlperf_resnet34_coco_1200_1200_433G	247.5	21.1
52	mlperf_resnet50	tf_mlperf_resnet50_imagenet_224_224_8.19G	247.5	763.6
53	refinedet	tf_refinedet_VOC_320_320_81.9G	247.5	108.2
54	refinedet_medical	tf_RefineDet-Medical_EDD_320_320_9.83G	247.5	641.2
55	pruned_rcan	tf_rcan_DIV2K_360_640_0.98_86.95G	220	60.3
56	resnet50	tf2_resnet50_imagenet_224_224_7.76G	247.5	788.1
57	Inception_v3	tf2_inceptionv3_imagenet_299_299_11.5G	247.5	568.7
58	2d-unet	tf2_2d-unet_nuclei_128_128_5.31G	247.5	1601.1
59	ERFNet	tf2_erfnet_cityscapes_512_1024_54G	247.5	75.5
60	ENet	pt_ENet_cityscapes_512_1024_8.6G	247.5	108.3
61	SemanticFPN	pt_SemanticFPN_cityscapes_256_512_10G	247.5	644.8
62	ResNet20-face	pt_facerec-resnet20_mixed_112_96_3.5G	247.5	1848.3
63	face quality	pt_face-quality_80_60_61.68M	247.5	23449.1
64	face_reid_large	pt_facereid-large_96_96_515M	247.5	10570.1
65	face_reid_small	pt_facereid-small_80_80_90M	247.5	27253.0
66	person_reid	pt_personreid-res50_market1501_256_128_5.4G	247.5	1229.0
67	person_reid	pt_personreid-res18_market1501_176_80_1.1G	247.5	5144.1
68	salsanext	pt_salsanext_semantic-kitti_64_2048_0.6_20.4G	247.5	174.2
69	FPN-R18 (light-weight)	pt_FPN-resnet18_covid19-seg_352_352_22.7G	247.5	313.3
70	2d-unet	pt_unet_chaos-CT_512_512_23.3G	247.5	91.2
71	salsanext_v2	pt_salsanextv2_semantic-kitti_64_2048_0.75_32G	220	36.4
72	PMG	pt_pmg_rp2k_224_224_2.28G	220	1334.6
-	Inception_v3	torchvision_inception_v3	220	499.6
-	resnet50	torchvision_resnet50	247.5	763.2
-	SqueezeNet	torchvision_squeezenet	247.5	2847.2

The following table lists the performance number including end-to-end throughput latency for each model on the Alveo U50 board with 1 DPUCAHX8L kernel running at 250Mhz in Gen3x4:

No.	Model	Name	DPU Frequency(MHz)	E2E throughput (fps) Multi Thread
1	resnet50	cf_resnet50_imagenet_224_224_7.7G	250	80.6
2	resnet18	cf_resnet18_imagenet_224_224_3.65G	250	297.0
3	Inception_v1	cf_inceptionv1_imagenet_224_224_3.16G	250	322.5
4	Inception_v2	cf_inceptionv2_imagenet_224_224_4G	250	185.2
5	Inception_v3	cf_inceptionv3_imagenet_299_299_11.4G	250	102.3
6	Inception_v4	cf_inceptionv4_imagenet_299_299_24.5G	250	52.7
7	Mobilenet_v2	cf_mobilenetv2_imagenet_224_224_0.59G	250	1018.2
8	SqueezeNet	cf_squeezenet_imagenet_227_227_0.76G	250	793.6
9	ssd_pedestrian_pruned_0_97	cf_ssdpedestrian_coco_360_640_0.97_5.9G	250	109.1
10	refinedet_baseline	cf_refinedet_coco_360_480_123G	250	26.3
11	refinedet_pruned_0_8	cf_refinedet_coco_360_480_0.8_25G	250	64.7
12	refinedet_pruned_0_92	cf_refinedet_coco_360_480_0.92_10.10G	250	76.3
13	refinedet_pruned_0_96	cf_refinedet_coco_360_480_0.96_5.08G	250	93.8
14	ssd_adas_pruned_0_95	cf_ssdadas_bdd_360_480_0.95_6.3G	250	118.3
15	ssd_traffic_pruned_0_9	cf_ssdtraffic_360_480_0.9_11.6G	250	111.2
16	VPGnet_pruned_0_99	cf_VPGnet_caltechlane_480_640_0.99_2.5G	250	74.1
17	ssd_mobilenet_v2	cf_ssdmobilenetv2_bdd_360_480_6.57G	250	138.0
18	FPN	cf_fpn_cityscapes_256_512_8.9G	250	38.8
19	SP_net	cf_SPnet_aichallenger_224_128_0.54G	250	10939.0
20	Openpose_pruned_0_3	cf_openpose_aichallenger_368_368_0.3_189.7G	250	14.4
21	face_landmark	cf_landmark_celeba_96_72_0.14G	250	4587.8
22	reid	cf_reid_market1501_160_80_0.95G	250	570.7
23	multi_task	cf_multitask_bdd_288_512_14.8G	250	21.1
24	ResNet20-face	cf_facerec-resnet20_112_96_3.5G	250	295.3
25	ResNet64-face	cf_facerec-resnet64_112_96_11G	250	129.1
26	FPN_Res18_segmentation	cf_FPN-resnet18_EDD_320_320_45.3G	250	14.2
27	plate detection	cf_plate-detection_320_320_0.49G	250	1105.0
28	retinaface	cf_retinaface_wider_360_640_1.11G	250	196.8
29	face_quality	cf_face-quality_80_60_61.68M	250	6557.5
30	Inception_resnet_v2	tf_inceptionresnetv2_imagenet_299_299_26.35G	250	32.2
31	Inception_v1	tf_inceptionv1_imagenet_224_224_3G	250	327.4
32	Inception_v3	tf_inceptionv3_imagenet_299_299_11.45G	250	99.7
33	Inception_v4	tf_inceptionv4_imagenet_299_299_24.55G	250	51.4
34	Mobilenet_v1	tf_mobilenetv1_0.5_imagenet_160_160_150.07M	250	4706.6
35	Mobilenet_v1	tf_mobilenetv1_1.0_imagenet_224_224_1.14G	250	1713.4
36	Mobilenet_v2	tf_mobilenetv2_1.0_imagenet_224_224_0.59G	250	1010.7
37	Mobilenet_v2	tf_mobilenetv2_1.4_imagenet_224_224_1.16G	250	745.0
38	resnet_v1_50	tf_resnetv1_50_imagenet_224_224_6.97G	250	93.4
39	resnet_v1_101	tf_resnetv1_101_imagenet_224_224_14.4G	250	52.6
40	resnet_v1_152	tf_resnetv1_152_imagenet_224_224_21.83G	250	35.8
41	vgg_16	tf_vgg16_imagenet_224_224_30.96G	250	52.2
42	vgg_19	tf_vgg19_imagenet_224_224_39.28G	250	46.1
43	ssd_mobilenet_v1	tf_ssdmobilenetv1_coco_300_300_2.47G	250	853.3
44	ssd_mobilenet_v2	tf_ssdmobilenetv2_coco_300_300_3.75G	250	210.3
45	ssdlite_mobilenetv2	tf_ssdlite_mobilenetv2_coco_300_300_1.5G	250	523.3
46	mlperf_ssd_resnet34	tf_mlperf_resnet34_coco_1200_1200_433G	250	6.2
47	mlperf_resnet50	tf_mlperf_resnet50_imagenet_224_224_8.19G	250	81.2
48	refinedet	tf_refinedet_VOC_320_320_81.9G	250	39.8
49	refinedet_medical	tf_RefineDet-Medical_EDD_320_320_9.83G	250	145.4
50	pruned_rcan	tf_rcan_DIV2K_360_640_0.98_86.95G	250	4.0
51	resnet50	tf2_resnet50_imagenet_224_224_7.76G	250	80.6
52	Mobilenet_v1	tf2_mobilenetv1_imagenet_224_224_1.15G	250	1683.1
53	Inception_v3	tf2_inceptionv3_imagenet_299_299_11.5G	250	94.2
54	2d-unet	tf2_2d-unet_nuclei_128_128_5.31G	250	113.9
55	ERFNet	tf2_erfnet_cityscapes_512_1024_54G	250	5.4
56	ENet	pt_ENet_cityscapes_512_1024_8.6G	250	5.7
57	SemanticFPN	pt_SemanticFPN_cityscapes_256_512_10G	250	35.7
58	ResNet20-face	pt_facerec-resnet20_mixed_112_96_3.5G	250	294.4
59	face quality	pt_face-quality_80_60_61.68M	250	6582.1
60	face_reid_small	pt_facereid-small_80_80_90M	250	4305.4
61	person_reid	pt_personreid-res50_market1501_256_128_5.4G	250	99.2
62	salsanext	pt_salsanext_semantic-kitti_64_2048_0.6_20.4G	250	11.6
63	FPN-R18 (light-weight)	pt_FPN-resnet18_covid19-seg_352_352_22.7G	250	82.1
64	SemanticFPN-mobilenetv2	pt_SemanticFPN-mobilenetv2_cityscapes_512_1024_5.4G	250	10.9
-	Inception_v3	torchvision_inception_v3	250	99.4
-	resnet50	torchvision_resnet50	250	82.6
-	SqueezeNet	torchvision_squeezenet	250	471.6

Performance on U200

Measured with Vitis AI 1.4 and Vitis AI Library 1.4

Click here to view details

The following table lists the performance number including end-to-end throughput and latency for each model on the Alveo U200 board with 2 DPUCADF8H kernels running at 300Mhz:

No.	Model	Name	E2E latency (ms) Thread num =20	E2E throughput -fps(Multi Thread)
1	resnet50	cf_resnet50_imagenet_224_224_7.7G	3.8	1054
2	Inception_v1	tf_inceptionv1_imagenet_224_224_3G	2.2	1834
3	Inception_v3	tf_inceptionv3_imagenet_299_299_11.45G	18.4	218
4	resnetv1_50	tf_resnetv1_50_imagenet_224_224_6.97G	4.2	947
5	resnetv1_101	tf_resnetv1_101_imagenet_224_224_14.4G	8.5	472
6	resnetv1_152	tf_resnetv1_152_imagenet_224_224_21.83G	12.7	316

The following table lists the performance number including end-to-end throughput and latency for each model on the Alveo U200 board with 2 DPUCADX8G kernels running at 350Mhz with xilinx_u200_xdma_201830_2 shell:

No.	Model	Name	E2E latency (ms) Thread num =1	E2E throughput -fps(Single Thread)	E2E throughput -fps(Multi Thread)
1	resnet50	cf_resnet50_imagenet_224_224_7.7G	2.13	470.6	561.3
2	resnet18	cf_resnet18_imagenet_224_224_3.65G	2.08	481	1157.8
3	Inception_v1	cf_inceptionv1_imagenet_224_224_3.16G	2.39	418.5	1449.4
4	Inception_v2	cf_inceptionv2_imagenet_224_224_4G	2.11	475.1	1129.2
5	Inception_v3	cf_inceptionv3_imagenet_299_299_11.4G	15.67	63.8	371.6
6	Inception_v4	cf_inceptionv4_imagenet_299_299_24.5G	10.77	92.8	221.2
7	SqueezeNet	cf_squeeze_imagenet_227_227_0.76G	10.99	91	1157.1
8	densebox_320_320	cf_densebox_wider_320_320_0.49G	8.69	115.1	667.9
9	yolov3_bdd	dk_yolov3_bdd_288_512_53.7G	14.53	68.8	75.9
10	yolov3_voc	dk_yolov3_voc_416_416_65.42G	19.90	50.3	82.1

Performance on U250

Measured with Vitis AI 1.4 and Vitis AI Library 1.4

Click here to view details

The following table lists the performance number including end-to-end throughput and latency for each model on the Alveo U250 board with 4 DPUCADF8H kernels running at 300Mhz:

No.	Model	Name	E2E latency (ms) Thread num =20	E2E throughput -fps(Multi Thread)
1	resnet50	cf_resnet50_imagenet_224_224_7.7G	1.94	2134.8
2	Inception_v1	tf_inceptionv1_imagenet_224_224_3G	1.10	3631.7
3	Inception_v3	tf_inceptionv3_imagenet_299_299_11.45G	9.20	434.9
4	resnetv1_50	tf_resnetv1_50_imagenet_224_224_6.97G	2.13	1881.6
5	resnetv1_101	tf_resnetv1_101_imagenet_224_224_14.4G	4.24	941.9
6	resnetv1_152	tf_resnetv1_152_imagenet_224_224_21.83G	6.35	630.3

The following table lists the performance number including end-to-end throughput and latency for each model on the Alveo U250 board with 4 DPUCADX8G kernels running at 350Mhz with xilinx_u250_xdma_201830_1 shell:

No.	Model	Name	E2E latency (ms) Thread num =1	E2E throughput -fps(Single Thread)	E2E throughput -fps(Multi Thread)
1	resnet50	cf_resnet50_imagenet_224_224_7.7G	1.68	595.5	1223.95
2	resnet18	cf_resnet18_imagenet_224_224_3.65G	1.67	600.5	2422.5
3	Inception_v1	cf_inceptionv1_imagenet_224_224_3.16G	1.93	517.1	4059.8
4	Inception_v2	cf_inceptionv2_imagenet_224_224_4G	1.65	607.8	23221
5	Inception_v3	cf_inceptionv3_imagenet_299_299_11.4G	6.18	161.8	743.8
6	Inception_v4	cf_inceptionv4_imagenet_299_299_24.5G	5.77	173.4	452.4
7	SqueezeNet	cf_squeeze_imagenet_227_227_0.76G	5.44	183.7	2349.7
8	densebox_320_320	cf_densebox_wider_320_320_0.49G	7.43	167.2	898.5
9	yolov3_bdd	dk_yolov3_bdd_288_512_53.7G	14.27	70.1	146.7
10	yolov3_voc	dk_yolov3_voc_416_416_65.42G	9.46	105.7	139.4

Note: For xilinx_u250_gen3x16_xdma_shell_3_1_202020_1 latest shell U250 xclbins, Alveo U250 board would be having only 3 DPUCADF8H kernels instead of 4, thereby the performance numbers for 1 Alveo U250 board with xilinx_u250_gen3x16_xdma_shell_3_1 shell xclbins would be 75% of the above reported performance numbers which is for 4 DPUCADF8H kernels.

Performance on U280

Measured with Vitis AI 1.4 and Vitis AI Library 1.4

Click here to view details

The following table lists the performance number including end-to-end throughput for each model on the Alveo U280 board with 14 DPUCAHX8H kernels running at 300Mhz:

No.	Model	Name	DPU Frequency(MHz)	E2E throughput (fps) Multi Thread
1	resnet50	cf_resnet50_imagenet_224_224_7.7G	150	800.1
2	resnet18	cf_resnet18_imagenet_224_224_3.65G	150	1658.0
3	Inception_v1	cf_inceptionv1_imagenet_224_224_3.16G	150	1378.7
4	Inception_v2	cf_inceptionv2_imagenet_224_224_4G	150	1100.4
5	Inception_v3	cf_inceptionv3_imagenet_299_299_11.4G	150	444.3
6	Inception_v4	cf_inceptionv4_imagenet_299_299_24.5G	150	208.4
7	SqueezeNet	cf_squeezenet_imagenet_227_227_0.76G	150	3961.0
8	ssd_pedestrian_pruned_0_97	cf_ssdpedestrian_coco_360_640_0.97_5.9G	150	576.0
9	refinedet_baseline	cf_refinedet_coco_360_480_123G	150	60.3
10	refinedet_pruned_0_8	cf_refinedet_coco_360_480_0.8_25G	150	220.2
11	refinedet_pruned_0_92	cf_refinedet_coco_360_480_0.92_10.10G	150	457.7
12	refinedet_pruned_0_96	cf_refinedet_coco_360_480_0.96_5.08G	150	646.0
13	ssd_adas_pruned_0_95	cf_ssdadas_bdd_360_480_0.95_6.3G	150	652.6
14	ssd_traffic_pruned_0_9	cf_ssdtraffic_360_480_0.9_11.6G	150	435.3
15	VPGnet_pruned_0_99	cf_VPGnet_caltechlane_480_640_0.99_2.5G	150	729.7
16	FPN	cf_fpn_cityscapes_256_512_8.9G	150	475.8
17	SP_net	cf_SPnet_aichallenger_224_128_0.54G	150	4036.3
18	Openpose_pruned_0_3	cf_openpose_aichallenger_368_368_0.3_189.7G	150	37.4
19	densebox_320_320	cf_densebox_wider_320_320_0.49G	150	4690.9
20	densebox_640_360	cf_densebox_wider_360_640_1.11G	150	2152.4
21	face_landmark	cf_landmark_celeba_96_72_0.14G	150	12488.2
22	reid	cf_reid_market1501_160_80_0.95G	150	4773.2
23	multi_task	cf_multitask_bdd_288_512_14.8G	150	353.1
24	yolov3_bdd	dk_yolov3_bdd_288_512_53.7G	150	94.0
25	yolov3_adas_pruned_0_9	dk_yolov3_cityscapes_256_512_0.9_5.46G	150	816.8
26	yolov3_voc	dk_yolov3_voc_416_416_65.42G	150	97.3
27	yolov2_voc	dk_yolov2_voc_448_448_34G	150	202.2
28	yolov2_voc_pruned_0_66	dk_yolov2_voc_448_448_0.66_11.56G	150	499.8
29	yolov2_voc_pruned_0_71	dk_yolov2_voc_448_448_0.71_9.86G	150	581.9
30	yolov2_voc_pruned_0_77	dk_yolov2_voc_448_448_0.77_7.82G	150	693.9
31	ResNet20-face	cf_facerec-resnet20_112_96_3.5G	150	1599.5
32	ResNet64-face	cf_facerec-resnet64_112_96_11G	150	579.2
33	FPN_Res18_segmentation	cf_FPN-resnet18_EDD_320_320_45.3G	150	117.3
34	plate detection	cf_plate-detection_320_320_0.49G	150	8251.1
35	plate recognition	cf_plate-recognition_96_288_1.75G	150	1497.6
36	face_quality	cf_face-quality_80_60_61.68M	150	28492.0
37	tiny-yolov3	dk_tiny-yolov3_416_416_5.46G	150	1040.3
38	yolov4	dk_yolov4_coco_416_416_60.1G	150	100.2
39	pruned_yolov4	dk_yolov4_coco_416_416_0.36_38.2G	150	123.5
40	Inception_resnet_v2	tf_inceptionresnetv2_imagenet_299_299_26.35G	150	192.4
41	Inception_v1	tf_inceptionv1_imagenet_224_224_3G	150	1431.4
42	Inception_v3	tf_inceptionv3_imagenet_299_299_11.45G	150	445.0
43	Inception_v4	tf_inceptionv4_imagenet_299_299_24.55G	150	208.4
44	resnet_v1_50	tf_resnetv1_50_imagenet_224_224_6.97G	150	749.8
45	resnet_v1_101	tf_resnetv1_101_imagenet_224_224_14.4G	150	389.0
46	resnet_v1_152	tf_resnetv1_152_imagenet_224_224_21.83G	150	259.7
47	vgg_16	tf_vgg16_imagenet_224_224_30.96G	150	188.6
48	vgg_19	tf_vgg19_imagenet_224_224_39.28G	150	157.5
49	ssd_resnet_50_v1_fpn	tf_ssdresnet50v1_fpn_coco_640_640_178.4G	150	37.2
50	yolov3_voc	tf_yolov3_voc_416_416_65.63G	150	97.2
51	mlperf_ssd_resnet34	tf_mlperf_resnet34_coco_1200_1200_433G	150	16.5
52	mlperf_resnet50	tf_mlperf_resnet50_imagenet_224_224_8.19G	150	645.7
53	refinedet	tf_refinedet_VOC_320_320_81.9G	150	87.9
54	refinedet_medical	tf_RefineDet-Medical_EDD_320_320_9.83G	150	514.8
55	pruned_rcan	tf_rcan_DIV2K_360_640_0.98_86.95G	150	51.5
56	resnet50	tf2_resnet50_imagenet_224_224_7.76G	150	666.3
57	Inception_v3	tf2_inceptionv3_imagenet_299_299_11.5G	150	454.9
58	2d-unet	tf2_2d-unet_nuclei_128_128_5.31G	150	1287.3
59	ERFNet	tf2_erfnet_cityscapes_512_1024_54G	150	67.1
60	ENet	pt_ENet_cityscapes_512_1024_8.6G	150	121.8
61	SemanticFPN	pt_SemanticFPN_cityscapes_256_512_10G	150	534.6
62	ResNet20-face	pt_facerec-resnet20_mixed_112_96_3.5G	150	1598.9
63	face quality	pt_face-quality_80_60_61.68M	150	29557.6
64	face_reid_large	pt_facereid-large_96_96_515M	150	11573.5
65	face_reid_small	pt_facereid-small_80_80_90M	150	30743.7
66	person_reid	pt_personreid-res50_market1501_256_128_5.4G	150	1042.6
67	person_reid	pt_personreid-res18_market1501_176_80_1.1G	150	4524.7
68	salsanext	pt_salsanext_semantic-kitti_64_2048_0.6_20.4G	150	108.5
69	FPN-R18 (light-weight)	pt_FPN-resnet18_covid19-seg_352_352_22.7G	150	263.7
70	2d-unet	pt_unet_chaos-CT_512_512_23.3G	150	132.4
71	salsanext_v2	pt_salsanextv2_semantic-kitti_64_2048_0.75_32G	150	55.0
72	PMG	pt_pmg_rp2k_224_224_2.28G	150	1270.9
-	Inception_v3	torchvision_inception_v3	150	445.2
-	resnet50	torchvision_resnet50	150	774.9
-	SqueezeNet	torchvision_squeezenet	150	2187.5

The following table lists the performance number including end-to-end throughput for each model on the Alveo U280 board with 2 DPUCAHX8L kernels running at 250Mhz:

No.	Model	Name	DPU Frequency(MHz)	E2E throughput (fps) Multi Thread
1	resnet50	cf_resnet50_imagenet_224_224_7.7G	250	160.5
2	resnet18	cf_resnet18_imagenet_224_224_3.65G	250	542.3
3	Inception_v1	cf_inceptionv1_imagenet_224_224_3.16G	250	639.6
4	Inception_v2	cf_inceptionv2_imagenet_224_224_4G	250	338.8
5	Inception_v3	cf_inceptionv3_imagenet_299_299_11.4G	250	189.1
6	Inception_v4	cf_inceptionv4_imagenet_299_299_24.5G	250	103.0
7	Mobilenet_v2	cf_mobilenetv2_imagenet_224_224_0.59G	250	2039.3
8	SqueezeNet	cf_squeezenet_imagenet_227_227_0.76G	250	1541.1
9	ssd_pedestrian_pruned_0_97	cf_ssdpedestrian_coco_360_640_0.97_5.9G	250	189.4
10	refinedet_baseline	cf_refinedet_coco_360_480_123G	250	51.9
11	refinedet_pruned_0_8	cf_refinedet_coco_360_480_0.8_25G	250	110.6
12	refinedet_pruned_0_92	cf_refinedet_coco_360_480_0.92_10.10G	250	129.1
13	refinedet_pruned_0_96	cf_refinedet_coco_360_480_0.96_5.08G	250	173.3
14	ssd_adas_pruned_0_95	cf_ssdadas_bdd_360_480_0.95_6.3G	250	199.5
15	ssd_traffic_pruned_0_9	cf_ssdtraffic_360_480_0.9_11.6G	250	202.6
16	VPGnet_pruned_0_99	cf_VPGnet_caltechlane_480_640_0.99_2.5G	250	114.2
17	ssd_mobilenet_v2	cf_ssdmobilenetv2_bdd_360_480_6.57G	250	231.6
18	FPN	cf_fpn_cityscapes_256_512_8.9G	250	49.9
19	SP_net	cf_SPnet_aichallenger_224_128_0.54G	250	13296.1
20	Openpose_pruned_0_3	cf_openpose_aichallenger_368_368_0.3_189.7G	250	28.1
21	face_landmark	cf_landmark_celeba_96_72_0.14G	250	9368.9
22	reid	cf_reid_market1501_160_80_0.95G	250	922.1
23	multi_task	cf_multitask_bdd_288_512_14.8G	250	26.6
24	ResNet20-face	cf_facerec-resnet20_112_96_3.5G	250	477.3
25	ResNet64-face	cf_facerec-resnet64_112_96_11G	250	228.3
26	FPN_Res18_segmentation	cf_FPN-resnet18_EDD_320_320_45.3G	250	16.8
27	plate detection	cf_plate-detection_320_320_0.49G	250	2278.8
28	retinaface	cf_retinaface_wider_360_640_1.11G	250	278.4
29	face_quality	cf_face-quality_80_60_61.68M	250	14078.7
30	Inception_resnet_v2	tf_inceptionresnetv2_imagenet_299_299_26.35G	250	64.9
31	Inception_v1	tf_inceptionv1_imagenet_224_224_3G	250	650.2
32	Inception_v3	tf_inceptionv3_imagenet_299_299_11.45G	250	188.3
33	Inception_v4	tf_inceptionv4_imagenet_299_299_24.55G	250	102.2
34	Mobilenet_v1	tf_mobilenetv1_0.5_imagenet_160_160_150.07M	250	9905.3
35	Mobilenet_v1	tf_mobilenetv1_1.0_imagenet_224_224_1.14G	250	3484.9
36	Mobilenet_v2	tf_mobilenetv2_1.0_imagenet_224_224_0.59G	250	2033.0
37	Mobilenet_v2	tf_mobilenetv2_1.4_imagenet_224_224_1.16G	250	1490.9
38	resnet_v1_50	tf_resnetv1_50_imagenet_224_224_6.97G	250	178.3
39	resnet_v1_101	tf_resnetv1_101_imagenet_224_224_14.4G	250	101.9
40	resnet_v1_152	tf_resnetv1_152_imagenet_224_224_21.83G	250	69.4
41	vgg_16	tf_vgg16_imagenet_224_224_30.96G	250	106.1
42	vgg_19	tf_vgg19_imagenet_224_224_39.28G	250	93.7
43	ssd_mobilenet_v1	tf_ssdmobilenetv1_coco_300_300_2.47G	250	1551.9
44	ssd_mobilenet_v2	tf_ssdmobilenetv2_coco_300_300_3.75G	250	374.0
45	ssdlite_mobilenetv2	tf_ssdlite_mobilenetv2_coco_300_300_1.5G	250	998.0
46	mlperf_ssd_resnet34	tf_mlperf_resnet34_coco_1200_1200_433G	250	12.4
47	mlperf_resnet50	tf_mlperf_resnet50_imagenet_224_224_8.19G	250	156.4
48	refinedet	tf_refinedet_VOC_320_320_81.9G	250	79.2
49	refinedet_medical	tf_RefineDet-Medical_EDD_320_320_9.83G	250	244.0
50	pruned_rcan	tf_rcan_DIV2K_360_640_0.98_86.95G	250	7.8
51	resnet50	tf2_resnet50_imagenet_224_224_7.76G	250	155.4
52	Mobilenet_v1	tf2_mobilenetv1_imagenet_224_224_1.15G	250	3423.9
53	Inception_v3	tf2_inceptionv3_imagenet_299_299_11.5G	250	180.6
54	2d-unet	tf2_2d-unet_nuclei_128_128_5.31G	250	136.3
55	ERFNet	tf2_erfnet_cityscapes_512_1024_54G	250	7.7
56	ENet	pt_ENet_cityscapes_512_1024_8.6G	250	8.0
57	SemanticFPN	pt_SemanticFPN_cityscapes_256_512_10G	250	43.7
58	ResNet20-face	pt_facerec-resnet20_mixed_112_96_3.5G	250	475.7
59	face quality	pt_face-quality_80_60_61.68M	250	14109.2
60	face_reid_small	pt_facereid-small_80_80_90M	250	8927.1
61	person_reid	pt_personreid-res50_market1501_256_128_5.4G	250	201.4
62	salsanext	pt_salsanext_semantic-kitti_64_2048_0.6_20.4G	250	21.2
63	FPN-R18 (light-weight)	pt_FPN-resnet18_covid19-seg_352_352_22.7G	250	128.5
64	SemanticFPN-mobilenetv2	pt_SemanticFPN-mobilenetv2_cityscapes_512_1024_5.4G	250	13.9
-	Inception_v3	torchvision_inception_v3	250	197.8
-	resnet50	torchvision_resnet50	250	156.8
-	SqueezeNet	torchvision_squeezenet	250	942.5

Performance on Ultra96

The performance number shown below was measured with the previous AI SDK v2.0.4 on Ultra96 v1. The Vitis platform of Ultra96 v2 has not been released yet. So the performance numbers are therefore not reported for this Model Zoo release.

Click here to view details

The following table lists the performance number including end-to-end throughput and latency for each model on the Ultra96 board with a 1 * B1600 @ 287MHz V1.4.0 DPU configuration:

Note: The original power supply of Ultra96 is not designed for high performance AI workload. The board may occasionally hang to run few models, When multi-thread is used. For such situations, NA is specified in the following table.

No.	Model	Name	E2E latency (ms) Thread num =1	E2E throughput -fps(Single Thread)	E2E throughput -fps(Multi Thread)
1	resnet50	cf_resnet50_imagenet_224_224_7.7G	30.8	32.4667	33.4667
2	Inception_v1	cf_inceptionv1_imagenet_224_224_3.16G	13.98	71.55	75.0667
3	Inception_v2	cf_inceptionv2_imagenet_224_224_4G	17.16	58.2667	61.2833
4	Inception_v3	cf_inceptionv3_imagenet_299_299_11.4G	44.05	22.7	23.4333
5	mobileNet_v2	cf_mobilenetv2_imagenet_224_224_0.59G	7.34	136.183	NA
6	tf_resnet50	tf_resnet50_imagenet_224_224_6.97G	28.02	35.6833	36.6
7	tf_inception_v1	tf_inceptionv1_imagenet_224_224_3G	16.96	58.9667	61.2833
8	tf_mobilenet_v2	tf_mobilenetv2_imagenet_224_224_1.17G	10.17	98.3	104.25
9	ssd_adas_pruned_0.95	cf_ssdadas_bdd_360_480_0.95_6.3G	24.3	41.15	46.2
10	ssd_pedestrian_pruned_0.97	cf_ssdpedestrian_coco_360_640_0.97_5.9G	23.29	42.9333	50.8
11	ssd_traffic_pruned_0.9	cf_ssdtraffic_360_480_0.9_11.6G	35.5	28.1667	31.8
12	ssd_mobilnet_v2	cf_ssdmobilenetv2_bdd_360_480_6.57G	60.79	16.45	27.8167
13	tf_ssd_voc	tf_ssd_voc_300_300_64.81G	186.92	5.35	5.81667
14	densebox_320_320	cf_densebox_wider_320_320_0.49G	4.17	239.883	334.167
15	densebox_360_640	cf_densebox_wider_360_640_1.11G	8.55	117	167.2
16	yolov3_adas_prune_0.9	dk_yolov3_cityscapes_256_512_0.9_5.46G	22.79	43.8833	49.6833
17	yolov3_voc	dk_yolov3_voc_416_416_65.42G	185.19	5.4	5.53
18	tf_yolov3_voc	tf_yolov3_voc_416_416_65.63G	199.34	5.01667	5.1
19	refinedet_pruned_0.8	cf_refinedet_coco_360_480_0.8_25G	66.37	15.0667	NA
20	refinedet_pruned_0.92	cf_refinedet_coco_360_480_0.92_10.10G	32.17	31.0883	33.6667
21	refinedet_pruned_0.96	cf_refinedet_coco_360_480_0.96_5.08G	20.29	49.2833	55.25
22	FPN	cf_fpn_cityscapes_256_512_8.9G	36.34	27.5167	NA
23	VPGnet_pruned_0.99	cf_VPGnet_caltechlane_480_640_0.99_2.5G	13.9	71.9333	NA
24	SP-net	cf_SPnet_aichallenger_224_128_0.54G	3.82	261.55	277.4
25	Openpose_pruned_0.3	cf_openpose_aichallenger_368_368_0.3_189.7G	560.75	1.78333	NA
26	yolov2_voc	dk_yolov2_voc_448_448_34G	118.11	8.46667	8.9
27	yolov2_voc_pruned_0.66	dk_yolov2_voc_448_448_0.66_11.56G	37.5	26.6667	30.65
28	yolov2_voc_pruned_0.71	dk_yolov2_voc_448_448_0.71_9.86G	30.99	32.2667	38.35
29	yolov2_voc_pruned_0.77	dk_yolov2_voc_448_448_0.77_7.82G	26.29	38.03333	46.8333
30	Inception-v4	cf_inceptionv4_imagenet_299_299_24.5G	88.76	11.2667	11.5333
31	SqueezeNet	cf_squeeze_imagenet_227_227_0.76G	5.96	167.867	283.583
32	face_landmark	cf_landmark_celeba_96_72_0.14G	2.95	339.183	347.633
33	reid	cf_reid_market1501_160_80_0.95G	6.28	159.15	166.633
34	yolov3_bdd	dk_yolov3_bdd_288_512_53.7G	193.55	5.16667	5.31667
35	tf_mobilenet_v1	tf_mobilenetv1_imagenet_224_224_1.14G	5.97	167.567	186.55
36	resnet18	cf_resnet18_imagenet_224_224_3.65G	13.47	74.2167	77.8167
37	resnet18_wide	tf_resnet18_imagenet_224_224_28G	97.72	10.2333	10.3833

Performance on ZCU102 (0432055-04)

This version of ZCU102 is out of stock. The performance number shown below was measured with the previous AI SDK v2.0.4. Now this form has stopped updating. So the performance numbers are therefore not reported for this Model Zoo release.

Click here to view details

The following table lists the performance number including end-to-end throughput and latency for each model on the ZCU102 (0432055-04) board with a 3 * B4096 @ 287MHz V1.4.0 DPU configuration:

No.	Model	Name	E2E latency (ms) Thread num =1	E2E throughput -fps(Single Thread)	E2E throughput -fps(Multi Thread)
1	resnet50	cf_resnet50_imagenet_224_224_7.7G	12.85	77.8	179.3
2	Inception_v1	cf_inceptionv1_imagenet_224_224_3.16G	5.47	182.683	485.533
3	Inception_v2	cf_inceptionv2_imagenet_224_224_4G	6.76	147.933	373.267
4	Inception_v3	cf_inceptionv3_imagenet_299_299_11.4G	17	58.8333	155.4
5	mobileNet_v2	cf_mobilenetv2_imagenet_224_224_0.59G	4.09	244.617	638.067
6	tf_resnet50	tf_resnet50_imagenet_224_224_6.97G	11.94	83.7833	191.417
7	tf_inception_v1	tf_inceptionv1_imagenet_224_224_3G	6.72	148.867	358.283
8	tf_mobilenet_v2	tf_mobilenetv2_imagenet_224_224_1.17G	5.46	183.117	458.65
9	ssd_adas_pruned_0.95	cf_ssdadas_bdd_360_480_0.95_6.3G	11.33	88.2667	320.5
10	ssd_pedestrian_pruned_0.97	cf_ssdpedestrian_coco_360_640_0.97_5.9G	12.96	77.1833	314.717
11	ssd_traffic_pruned_0.9	cf_ssdtraffic_360_480_0.9_11.6G	17.49	57.1833	218.183
12	ssd_mobilnet_v2	cf_ssdmobilenetv2_bdd_360_480_6.57G	24.21	41.3	141.233
13	tf_ssd_voc	tf_ssd_voc_300_300_64.81G	69.28	14.4333	46.7833
14	densebox_320_320	cf_densebox_wider_320_320_0.49G	2.43	412.183	1416.63
15	densebox_360_640	cf_densebox_wider_360_640_1.11G	5.01	199.717	719.75
16	yolov3_adas_prune_0.9	dk_yolov3_cityscapes_256_512_0.9_5.46G	11.09	90.1667	259.65
17	yolov3_voc	dk_yolov3_voc_416_416_65.42G	70.51	14.1833	44.4
18	tf_yolov3_voc	tf_yolov3_voc_416_416_65.63G	70.75	14.1333	44.0167
19	refinedet_pruned_0.8	cf_refinedet_coco_360_480_0.8_25G	29.91	33.4333	109.067
20	refinedet_pruned_0.92	cf_refinedet_coco_360_480_0.92_10.10G	15.39	64.9667	216.317
21	refinedet_pruned_0.96	cf_refinedet_coco_360_480_0.96_5.08G	11.04	90.5833	312
22	FPN	cf_fpn_cityscapes_256_512_8.9G	16.58	60.3	203.867
23	VPGnet_pruned_0.99	cf_VPGnet_caltechlane_480_640_0.99_2.5G	9.44	105.9	424.667
24	SP-net	cf_SPnet_aichallenger_224_128_0.54G	1.73	579.067	1620.67
25	Openpose_pruned_0.3	cf_openpose_aichallenger_368_368_0.3_189.7G	279.07	3.58333	38.5
26	yolov2_voc	dk_yolov2_voc_448_448_34G	39.76	25.15	86.35
27	yolov2_voc_pruned_0.66	dk_yolov2_voc_448_448_0.66_11.56G	18.42	54.2833	211.217
28	yolov2_voc_pruned_0.71	dk_yolov2_voc_448_448_0.71_9.86G	16.42	60.9167	242.433
29	yolov2_voc_pruned_0.77	dk_yolov2_voc_448_448_0.77_7.82G	14.46	69.1667	286.733
30	Inception-v4	cf_inceptionv4_imagenet_299_299_24.5G	34.25	29.2	84.25
31	SqueezeNet	cf_squeeze_imagenet_227_227_0.76G	3.6	277.65	1080.77
32	face_landmark	cf_landmark_celeba_96_72_0.14G	1.13	885.033	1623.3
33	reid	cf_reid_marketcuhk_160_80_0.95G	2.67	375	773.533
34	yolov3_bdd	dk_yolov3_bdd_288_512_53.7G	73.89	13.5333	42.8833
35	tf_mobilenet_v1	tf_mobilenetv1_imagenet_224_224_1.14G	3.2	312.067	875.967
36	resnet18	cf_resnet18_imagenet_224_224_3.65G	5.1	195.95	524.433
37	resnet18_wide	tf_resnet18_imagenet_224_224_28G	33.28	30.05	83.4167

Contributing

We welcome community contributions. When contributing to this repository, first discuss the change you wish to make via:

• GitHub Issues
• Forum
• Email

You can also submit a pull request with details on how to improve the product. Prior to submitting your pull request, ensure that you can build the product and run all the demos with your patch. In case of a larger feature, provide a relevant demo.

License

Xilinx AI Model Zoo is licensed under Apache License Version 2.0. By contributing to the project, you agree to the license and copyright terms therein and release your contribution under these terms.

---

^{Copyright© 2019 Xilinx}