This work applies FSSL and SSL on the Zenseact Open Dataset (ZOD) for the detection of other cars.
Steps to execute:
- Prepare resized bounding box data by running prepare_df_boundingboxes.py
- Train SSL or FSSL model via train_simsiam.py or train_fedsimsiam.py
- Train the Faster R-CNN which uses the trained backbone weights from the previous step by executing train_fasterrcnn.py
- Receive the AP score by running test_fasterrcnn.py
The following figure shows the labeled bounding boxes on the left, and the predicted bounding boxes of a Faster R-CNN network with a backbone network trained via FSSL on the right.
