/camil_pytorch

Primary LanguagePython

TEST BED FOR PATCH MERGING:

About test bed:

Todo before running main file for test bed:

  1. activate conda/pip env and then:
    export PROJECT_DIR=$(pwd)
  2. Fix your path in testbed_config/{yourname_expnumber}.yaml

Main file:

python main_testbed_slide.py -> loop through all slide and return the patch (image) in the whole slide.

python main_testbed_superpixel.py -> loop through each slide, then loop through all superixel and return the patch (image) in the whole slide.

Camil Training:

----- Below config is to run Camil before start env export DATA_DIR=/project/hnguyen2/mvu9/camelyon16_features_data/h5_files

down load this file !gdown 1CS7I0yrTSNLbFk_CzqLrh5TKesZo3uXm then unzip them into data/camelyon16_feature/h5_files

  • to running the training: python train.py
  • to dry run (testing the code with few sample), run: python train.py --dry_run True
.
├── README.md
├── check_cuda.py
├── data
│   ├── camelyon16_dataset.py
│   ├── camelyon16_features
│   │   └── h5_files
│   ├── camelyon_csv_splits
│   │   ├── splits_0.csv
│   │   ├── splits_1.csv
│   │   ├── splits_2.csv
│   │   ├── splits_3.csv
│   │   └── splits_4.csv
│   ├── label_files
│   │   ├── camelyon_17.csv
│   │   ├── camelyon_data.csv
│   │   └── tcga_data.csv
│   ├── logs
│   └── weights
├── feature_extractor
├── requirements.txt
├── scripts
├── src
│   ├── __init__.py
│   ├── camil.py
│   ├── custom_layers.py
│   ├── nystromformer.py
├── train.py
└── utils
    ├── __init__.py
    ├── eval.py
    ├── helper.py
    └── utils.py

Experiment 01:

  • use pretrained embedding
  • learning rate: 1e-05
  • epochs: 30

image