imagenet_localization

This repo contains Python code for generating bounding box files which can be used to compute ImageNet localization results.

0. Prerequisites.

Run pip install -r requirements.txt to install the needed packages.

1. Generate bounding box file with predicted labels and bounding boxes.

Run python generate_bbox_file.py DATA_DIR OUT_FILE, where DATA_DIR is the path to the directory containing torch result files and OUT_FILE is the path for the output file.

2. Compute localization results.

Run python compute_localization_results.py --bb_file BB_FILE --annotation_di ANNOTATION_DIR, where BB_FILE is the path to the output file from the first steph and ANNOTATION_DIR is the path to the directory containing ImageNet XML annotation files.