Experimental code for a master thesis on client-side filtering for continuous learning from streaming perception data in a federated setting. For each incoming stream item, the system decides whether to train, store for replay, or skip, using model-driven selection criteria.
From the repo root:
# Offline classification baseline (upper bound)
python experiments/offline_classification.py --config configs/classification/offline.yaml
# Streaming classification with replay
python experiments/streaming_classification.py --config configs/classification/streaming_no_filter.yamlDetection experiments use experiments/streaming_detection.py and configs/detection/streaming_*.yaml.
Dependencies are in pyproject.toml. From the repo root:
pip install -e .Early experimental. Codebase and configs are under active development.
See LICENSE.