pit.md

PiT

• 论文：Rethinking Spatial Dimensions of Vision Transformers

• 官方项目：naver-ai/pit

• 模型代码：pit.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])
  ])

• 模型细节：

  Model	Model Name	Params (M)	FLOPs (G)	Top-1 (%)	Top-5 (%)	Pretrained Model
  PiT-Ti	pit_ti	4.9	0.7	72.91	91.40	Download
  PiT-XS	pit_xs	10.6	1.4	78.18	94.16	Download
  PiT-S	pit_s	23.5	2.9	81.08	95.33	Download
  PiT-B	pit_b	73.8	12.5	82.44	95.71	Download
  PiT-Ti distilled	pit_ti_distilled	4.9	0.7	74.54	92.10	Download
  PiT-XS distilled	pit_xs_distilled	10.6	1.4	79.31	94.36	Download
  PiT-S distilled	pit_s_distilled	23.5	2.9	81.99	95.79	Download
  PiT-B distilled	pit_b_distilled	73.8	12.5	84.14	96.86	Download

• 引用：

  @article{heo2021pit,
      title={Rethinking Spatial Dimensions of Vision Transformers},
      author={Byeongho Heo and Sangdoo Yun and Dongyoon Han and Sanghyuk Chun and Junsuk Choe and Seong Joon Oh},
      journal={arXiv: 2103.16302},
      year={2021},
  }