/CHOLEC80-CVS-PUBLIC

Primary LanguagePythonMIT LicenseMIT

COLENET: Baseline model for the Cholec80-CSV dataset

License: MIT

This repository holds the data preprocessing and the baseline model training and testing scripts for the dataset Cholec80-CSV, the first open dataset for Strasgerg's Criteria Detection.

USAGE

Download Cholec80-CVS dataset

Download Cholec80-CSV from its official repository.
Then store it on data/ with the name surgeons_annotations.xlsx

Configuration file

To ease the use of this repository we provide a configuration file. It must be filled before running any of the provided scripts.

  • dataset_path: Absolute path of CHOLEC80 dataset.
  • frames_path: Path where frames will be stored.

Data Wrangling

All the scripts inside the folder data_preprocessing transform the raw data into usable data formats. These scripts handle both the annotations provided by the surgeons and also the videos of the CHOLEC80 dataset. They must be executed from the data_preprocessing folder and must be executed in the following order:

  • get_valid_frames.py
  • video_2_frames.py
  • annotation_2_labels.py
  • get_training_sets.py

After that, the relevant frames of each video will be stored in the path specified on the config file, and also the training, validation and testing sets will be created on the data folder.

Model training

Regular training

To train any of the supported backbones simply run train.py. In this file, you can specify the backbone, and the log directory name. The script will automatically create a directory and store the training logs, and the best model weights. The supported backbones are:

vgg resnet alexnet densenet inception

Folds Training

If desired, it is possible to execute k-fold cross validation by executing the script train_folds.py as described in the dataset description paper. This script, automatically load the file data/folds.json which stores the splits used to generate the results reported.

Files Description

data_preprocessing/video_2_frames.py

This script divides the CHOLEC80 videos into separated frames, these frames are stored on the location specified on the config file. This procedure is executed on the frames that occurs before the Clipping and Cutting Phase.

data_preprocessing/get_valid_frames.py

This script extracts the index of the last valid frame, which corresponds with the last frame before the Clipping and Cutting Phase.

data_preprocessing/annotations_to_labels.py

This script transforms the annotations provided by the surgeons into a format easier to use in subsequent steps. It stores the annotations per video in the data/surgeons_annotations folder.