COCO val2017 mini-500

COCO val2017 mini-500 A frozen, reproducible 500-image subset of COCO val2017 for fast object-detection latency/throughput benchmarking on edge devices (e.g. LibreYOLO on NVIDIA Jetson), where the full 5000-image val set is impractical. Selection (deterministic) The 500 images are sorted(COCO.getImgIds()) :500 of the official instances val2017.json (the 500 lowest image IDs) — identical to running the vision-analysis-benchmark harness with –limit 500.… See the full description on the dataset page:

Source: Hugging Face Hub (LibreYOLO/coco-val2017-mini500). Metadata imported from the dataset’s Hub tags.