🩺 SegCol Challenge 2024

🌟 Semantic Segmentation for Tools and Fold Edges in Colonoscopy data

🌐 Challenge Website https://www.synapse.org/#!Synapse:syn54124209

📧 Contact: [email protected]

📈 Baseline

We have included Dexined and DeepLab as baselines for folds edge detection and tool segmentation, respectively. After generating the .npy predictions, the output from both methods can be merged using merge.py and then evaluated as done in Evaluation

We will use this two baseline models for Task2 evaluation. For each model, please refer to the instructions under the corresponding folders.

📋 Model Performance

Task 1 baseline scores are shown below.

Classes	ODS	OIS	Dice	AP	CLDice	Threshold
Folds (Dexined)	0.2461	0.2525	0.2684	0.0894	0.2736	0.82
Tool 1 (Deeplab)	0.0546	0.0721	0.3636	0.1330	0.2440	0.65
Tool 2 (Deeplab)	0.0489	0.0698	0.0930	0.0331	0.0414	0.37
Tool 3 (Deeplab)	0.1214	0.1255	0.8041	0.6515	0.2837	0.64

📊 Evaluation

Task 1 and 2 will be evaluated according to the eval.py script.

📝 Task 1

The evaluation for Task 1 requires firstly generating predictions according to 'docker_templates'.

1️⃣ Generate predictions:

cd docker_templates/Task1_dummy_docker
docker build -t <image_name> . 
docker run --gpus all -it --rm -v "$(pwd)/../../data:/data" <image_name> /data/input /data/output

2️⃣ Evaluate:

cd ../..
python eval.py

📝 Task 2

Task 2 requires firstly generating a text file according to 'docker_templates', then checking if the text file content format is correct:

1️⃣ Generate sample list:

cd docker_templates/Task2_dummy_docker
docker build -t <image_name> . 
docker run --gpus all -it --rm -v "$(pwd)/../../data:/data" <image_name> /data/input /data/output

2️⃣ Format check:

cd ../..
python Task2_file_format_check.py

The generated list of images with the training data will be used to retrain the baseline model and after training we will evaluate using eval.py.

3️⃣ Active learning strategy:

We provide an example of active selection for Task 2 in folder Active_Learning. Even though this example is based on DeepLab, in Task2 we will evaluate on both DexiNed and DeepLab for Folds and Tools respectively by training on the selected data list.