Verify evals on Papers with Code

Current status — 2026-09-05

The provider's comments report that the two unofficial paper entries were removed and that separate end-to-end/non-end-to-end leaderboard entries are not currently offered. Preserve that provider decision; duplicating rows is not automatically a better comparison.

The remaining task is to verify the 20 model-result rows and evaluation protocol against the official model tables and paper. The current public paper page was accessible in this audit, but its result-row details were not exposed in the retrieved page, so those rows are not live-verified. Repository model-table values alone do not establish that the provider imported them correctly. The earlier suggestion of a public API was explicitly corrected by the provider; do not build an integration against an assumed endpoint.

Smallest next step: obtain the actual result-row export or provider-visible values, compare model/task/dataset/metric/protocol one by one, and request only demonstrated corrections. Keep this metadata/provider tracker open; no package implementation change is indicated.

---

Original report and historical context

Hi,

Niels here from the open-source team at Hugging Face. Congratulations on your work!

I've made the paper and 20 verified paper-native evaluations available on Papers with Code.

The paper has results on Image Classification, Image segmentation, Object Detection, and 1 additional task page.

Would it be possible to verify these results and let me know if any score, model name, benchmark protocol, or openness metadata should be corrected? The imported rows are tied to the paper or its official release artifacts; comparison-table baselines were not added.

You can also edit the task, methods, project page, and GitHub URL directly from the paper page using your Hugging Face account.

Kind regards,

Niels