uncertainty-aware CADx
download the pretrained weight of mpvit-small
wget https://dl.dropbox.com/s/y3dnmmy8h4npz7a/mpvit_small.pth
python -m torch.distributed.launch --nproc_per_node 8 --use_env main.py --model mpvit_small --pretrained_mpvit <path/to/pretrained_ckpt> --data-path <path/to/polyp> --output <path/to/output_dir>
python -m torch.distributed.launch --nproc_per_node 8 --use_env main.py --model deit_small_patch16_224 --pretrained --data-path <path/to/polyp> --output <path/to/output_dir>
python -m torch.distributed.launch --nproc_per_node 8 --use_env main.py --model vit_small_patch16_224 --pretrained --data-path <path/to/polyp> --output <path/to/output_dir>
| Model | MPViT-Small | ViT-Small | DeiT-Small | ResNet-50 |
|---|---|---|---|---|
| Model size | 22.8M | 22.1M | 22.1M | 25.6M |
| Accuracy (Last top-1/ Max top-1) | 97.0 / 96.0 | 93.7 / 92.0 | 93.0 / 92.3 | 95.0 / 94.3 |
This repository is built on This repository is built using the Timm and MPViT for backbone networks and Transformer-Explainability for visualizing the attention map.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent