Datasets:
RipBench v1.2.2 - evaluation scripts (minimal upload)
eval/ holds the released evaluators of the RipBench paper, unchanged: evaluate_classification.py, evaluate_detection.py (axis-aligned boxes, mask-derived boxes and instance masks), evaluate_obb.py and evaluate_semantic.py, with results_io.py and video_metadata.py as their helpers. Each script states its inputs (prediction dumps + the labels/ views) and the frozen protocol (operating thresholds selected on the validation split, per-frame scoring) in its docstring; pip install -r requirements_tooling.txt.
The test split carries no annotations in this release, so the evaluators score the train and validation splits locally; test-split evaluation is done through the benchmark's submission procedure (see the dataset README). Table builders, per-video bootstrap tooling, the label and video-view builders and the result files of the paper are not part of this minimal upload.