cait.md

CaiT

• 论文：Going deeper with Image Transformers

• 官方项目：facebookresearch/deit

• 模型代码：cait.py

• 验证集数据处理：

  # 图像后端：pil
  # 输入图像大小：224x224
  transforms = T.Compose([
      T.Resize(248, interpolation='bicubic'),
      T.CenterCrop(224),
      T.ToTensor(),
      T.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
  ])
  
  # 图像后端：pil
  # 输入图像大小：384X384
  transforms = T.Compose([
      T.Resize(384, interpolation='bicubic'),
      T.CenterCrop(384),
      T.ToTensor(),
      T.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
  ])
  
  # 图像后端：pil
  # 输入图像大小：448X448
  transforms = T.Compose([
      T.Resize(448, interpolation='bicubic'),
      T.CenterCrop(448),
      T.ToTensor(),
      T.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
  ])

• 模型细节：

  Model	Model Name	Params (M)	FLOPs (G)	Top-1 (%)	Top-5 (%)	Pretrained Model
  CaiT-xxs-24	cait_xxs_24	12.0	2.5	78.50	94.30	Download
  CaiT-xxs-36	cait_xxs_36	17.3	3.8	79.87	94.90	Download
  CaiT-s-24	cait_s_24	49.6	9.4	83.40	96.62	Download
  CaiT-xxs-24-384	cait_xxs_24_384	12.0	9.5	80.97	95.64	Download
  CaiT-xxs-36-384	cait_xxs_36_384	17.3	14.2	82.20	96.15	Download
  CaiT-xs-24-384	cait_xs_24_384	26.6	19.3	84.06	96.89	Download
  CaiT-s-24-384	cait_s_24_384	49.6	32.2	85.05	97.34	Download
  CaiT-s-36-384	cait_s_36_384	68.2	48.0	85.45	97.48	Download
  CaiT-m-36-384	cait_m_36_384	270.9	173.3	86.06	97.73	Download
  CaiT-m-48-448	cait_m_48_448	356.0	329.6	86.49	97.75	Download

• 引用：

  @article{touvron2021cait,
  title={Going deeper with Image Transformers},
  author={Hugo Touvron and Matthieu Cord and Alexandre Sablayrolles and Gabriel Synnaeve and Herv'e J'egou},
  journal={arXiv preprint arXiv:2103.17239},
  year={2021}
  }