Densely-packed Object Detection via Hard Negative-Aware Anchor Attention

An official code of Densely-packed Object Detection via Hard Negative-Aware Anchor Attention in WACV2022

Paper : https://openaccess.thecvf.com/content/WACV2022/papers/Cho_Densely-Packed_Object_Detection_via_Hard_Negative-Aware_Anchor_Attention_WACV_2022_paper.pdf

Dataset

• SKU110k dataset : https://github.com/eg4000/SKU110K_CVPR19
• download ,unzip and build a structure like bellows

SUK110K_fixed (root)
            |-- SUK110K_fixed
                            |-- images
                                    |-- test_0.jpg
                                    |-- ...
                            |-- annotations
                                    |-- annotations_test.csv
                                    |-- ...
                            |-- LICENSE.txt

Pretrained file

• Download 'pretrained_model.pth.tar' from here (about 168MB)
• make './saves' folder and put it to './saves' folder.

Quick Start

• Download dataset
• Download pretrained_model

# python test.py
usage: test.py [-h] [--data_root]

  -h, --help            show this help message and exit
  --data_root           for testing, set your sku110k root path (default='D:\SKU110K_fixed')

Results

methods	Dataset	Resolution	AP	AP50	AP75	AR300	AR300^0.50	P(R=.5)
ours	SKU test	800~1333	0.522	0.897	0.556	0.601	0.935	0.816

Citation

@InProceedings{Cho_2022_WACV,
    author    = {Cho, Sungmin and Paeng, Jinwook and Kwon, Junseok},
    title     = {Densely-Packed Object Detection via Hard Negative-Aware Anchor Attention},
    booktitle = {Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)},
    month     = {January},
    year      = {2022},
    pages     = {2635-2644}
}