Skip to content

Latest commit

 

History

History

classification

README.md

Usage

Requirement:

  • PyTorch 1.8.0+
  • Python3.8
  • CUDA 10.1+
  • timm==0.4.5
  • tlt==0.1.0
  • pyyaml
  • apex-amp

CoTNet Results on ImageNet

name resolution #params FLOPs Top-1 Acc. Top-5 Acc. model
CoTNet-50 224 22.2M 3.3 81.3 95.6 GoogleDrive / Baidu
CoTNeXt-50 224 30.1M 4.3 82.1 95.9 GoogleDrive / Baidu
SE-CoTNetD-50 224 23.1M 4.1 81.6 95.8 GoogleDrive / Baidu
CoTNet-101 224 38.3M 6.1 82.8 96.2 GoogleDrive / Baidu
CoTNeXt-101 224 53.4M 8.2 83.2 96.4 GoogleDrive / Baidu
SE-CoTNetD-101 224 40.9M 8.5 83.2 96.5 GoogleDrive / Baidu
SE-CoTNetD-152 224 55.8M 17.0 84.0 97.0 GoogleDrive / Baidu
SE-CoTNetD-152 320 55.8M 26.5 84.6 97.1 GoogleDrive / Baidu

Access code for Baidu is cotn

Wave-ViT

Train

python3 -m torch.distributed.launch \
   --nproc_per_node=8 \
   --nnodes=1 \
   --node_rank=0 \
   --master_addr="localhost" \
   --master_port=12346 \
   --use_env main.py --config configs/wavevit/wavevit_s.py --data-path /export/home/dataset/imagenet --epochs 310 --batch-size 128 \
   --token-label --token-label-size 7 --token-label-data /export/home/dataset/imagenet/label_top5_train_nfnet

Results on ImageNet

name resolution #params FLOPs Top-1 Acc. Top-5 Acc. model
Wave-ViT-S 224 22.7M 4.7 83.9 96.6 GoogleDrive / Baidu
Wave-ViT-S 384 22.7M 15.1 85.0 97.2 GoogleDrive / Baidu
Wave-ViT-B 224 33.5M 7.2 84.8 97.1 GoogleDrive / Baidu
Wave-ViT-B 384 33.5M 23.4 85.6 97.4 GoogleDrive / Baidu
Wave-ViT-L 224 57.5M 14.8 85.5 97.3 GoogleDrive / Baidu
Wave-ViT-L 384 57.5M 47.5 86.3 97.7 GoogleDrive / Baidu

Access code for Baidu is nets

Dual-ViT

Train

python3 -m torch.distributed.launch \
   --nproc_per_node=8 \
   --nnodes=1 \
   --node_rank=0 \
   --master_addr="localhost" \
   --master_port=12346 \
   --use_env main.py --config configs/dualvit/dualvit_s.py --data-path /export/home/dataset/imagenet --epochs 310 --batch-size 128 \
   --token-label --token-label-size 7 --token-label-data /export/home/dataset/imagenet/label_top5_train_nfnet

Results on ImageNet

name resolution #params FLOPs Top-1 Acc. Top-5 Acc. model
Dual-ViT-S 224 25.1M 5.4 84.1 96.8 Baidu
Dual-ViT-S 384 25.1M 18.4 85.2 97.3 Baidu
Dual-ViT-B 224 42.6M 9.3 85.2 97.2 Baidu
Dual-ViT-B 384 42.6M 31.9 86.0 97.6 Baidu
Dual-ViT-L 224 73.0M 18.0 85.7 97.4 Baidu
Dual-ViT-L 384 73.0M 60.5 86.5 97.8 Baidu

Access code for Baidu is nets