Skip to content

Latest commit

 

History

History
77 lines (54 loc) · 3.61 KB

GetStarted.md

File metadata and controls

77 lines (54 loc) · 3.61 KB
Raw

Usage

Note: Currently, for video-based methods, one GPU could only hold 1 image. Do not put 2 or more images on 1 GPU!

Data preparation

Please download CVC-Clinic and ASUMayo endoscopic video datasets. After that, we recommend converting the mask images to PASCAL VOC XML annotations (scripts refer to mask_to_voc.py) and symlinking the converted data path to datasets/. The path structure should be as follows:

STFT
├── datasets
│   ├── ASUVideo
│   │   ├── Annotations
│   │   │   ├── subdir-annos
│   │   ├── Data
│   │   │   ├── subdir-images
│   │   ├── ImageSets
│   │   │   ├── ASUVideo_train_videos.txt
│   │   │   ├── ASUVideo_val_videos.txt
│   ├── CVCVideo
│   │   ├── Annotations
│   │   │   ├── subdir-annos
│   │   ├── Data
│   │   │   ├── subdir-images
│   │   ├── ImageSets
│   │   │   ├── CVCVideo_train_videos.txt
│   │   │   ├── CVCVideo_val_videos.txt

Note: Since the two databases are protected by copyright, we only uploaded the preprocessed data of one subdirectory in the test dataset at drive for data references and model inference. In order to quickly test our model, we have already provided the image list of this subdirectory as a txt file under directory datasets/ASUVideo/ImageSets. If you are interested in using these databases, please contact the copyright owner.

Inference

The inference command line for testing on the ASUMayo validation set with 1 GPU:

python -m torch.distributed.launch \
    --nproc_per_node=1 \
    tools/test_net.py \
    --master_port=$((RANDOM + 10000)) \
    --config-file configs/STFT/asuvid_R_50_STFT.yaml \
    MODEL.WEIGHT pretrained_models/ASUMayo_STFT_R_50.pth \
    OUTPUT_DIR log_dir/asuvid_R_50_STFT \
    TEST.IMS_PER_BATCH 1

Please note that:

  1. 1 GPU only holds 1 images for STFT, you should keep TEST.IMS_PER_BATCH equal to the number of GPUs you use.
  2. The pretrained model of STFT on the ASUMayo dataset is available at here. After downloaded, it should be placed at pretrained_models/.
  3. If you want to record the detailed results, please specify OUTPUT_DIR. Meanwhile, you can visualize the test results by adding --visulize option.
  4. If you want to evaluate a different model, please change --config-file and MODEL.WEIGHT.

Training

The following command line will train asuvid_R_50_STFT on 4 GPUs:

python -m torch.distributed.launch \
    --nproc_per_node=4 \
    tools/train_net.py \
    --master_port=$((RANDOM + 10000)) \
    --config-file configs/STFT/asuvid_R_50_STFT.yaml \
    OUTPUT_DIR log_dir/asuvid_R_50_STFT

Please note that:

  1. The models will be saved into OUTPUT_DIR.
  2. If you want to train other methods, please change --config-file.
  3. For training FGFA, the pretrained weight of FlowNet is available at here. After downloaded, it should be placed at pretrained_models/.

Customize

If you want to use these methods on your own dataset or implement your new method. Please refer to here.