Active Learning for Semantic Segmentation

Installation

Install dependencies:

pip install -r requirements.txt

Adjust parameters in src/params.py.

Simple train loop

Currently supproted datasets:

• CamVid
• BDD
• Cityscapes

python src/train_semantic_segmentation.py

Active learning

Currently implemented anomaly detection strategies:

• Entropy sampling,
• Marging sampling,
• Query-by-committee.

Entropy and Margin samples selection:

python src/entropy_margin.py

Query-by-committee samples selection:

python src/query_by_committee.py

Tensorboard logging

Start tensorboard, specifying the path to logging directory.

tensorboard --logdir src/tb_runs/EXPERIMENT_NAME

In your browser open http://localhost:6006 to see the results.

Reference:

• J. Prendki, An Introduction to Active Learning, ODSC
• Active Learning Tutorial
• B. Settles, Active Learning Literature Survey, CS Technical Report
• Presentation