| name | benchmark-on-visionanalysis |
| description | Signpost for benchmarking LibreYOLO models for visionanalysis.org. Use when someone wants to "benchmark for visionanalysis", produce a submission for the site, measure a model on COCO for publication, or add a hardware/runtime row to the site. The actual work lives in two OTHER repos; this skill orients you and hands off. It does not run benchmarks itself. |
Benchmark a LibreYOLO model for visionanalysis.org
This is a pointer skill. The benchmark and publish logic live in two separate
repos, each with its own authoritative skill. Do not run benchmarks from inside
libreyolo; switch to the right repo and follow its skill.
The two repos
Direct links to the authoritative skills (read these — this signpost only orients):
Local checkouts on this machine:
C:\Users\Usuario\Documents\GitHub\vision-analysis-benchmark and
C:\Users\Usuario\Documents\GitHub\vision-analysis.
GitHub: LibreYOLO/vision-analysis-benchmark, LibreYOLO/vision-analysis.
What to know before you start
The dataset, layout gotchas, protocol config, supported runtimes, and the
libreyolo_commit rule all live in the harness skill's "Dataset & protocol"
section (generate-benchmark-results). Read that there; this signpost does not
copy it (so it cannot drift). One thing worth knowing up front: the canonical
eval set is the HF dataset LibreYOLO/coco-val2017-mini500, not full COCO.
Flow
- Go to
vision-analysis-benchmark, follow generate-benchmark-results. This emits
one va.submission.v1 JSON per (model x runtime x hardware) run.
- Hand the emitted JSON(s) to
vision-analysis, follow submit-benchmark-results
to validate, rebuild generated/verified-results.v1.json, and open the PR / deploy.
The two skills above are authoritative. If anything here disagrees with them, they
win - update this signpost rather than diverging.
