Verify evals on Papers with Code

Hi,

Niels here from the open-source team at Hugging Face.

I've made the following papers and their evaluation results available on Papers with Code:

- [Perception Encoder: The best visual embeddings are not at the output of the network](https://paperswithcode.co/paper/2504.13181) — [38 paper-native evaluations](https://paperswithcode.co/paper/2504.13181#results).
- [PerceptionLM: Open-Access Data and Models for Detailed Visual Understanding](https://paperswithcode.co/paper/2504.13180) — [24 paper-native evaluations](https://paperswithcode.co/paper/2504.13180#results).
- [Pushing the Frontier of Audiovisual Perception with Large-Scale Multimodal Correspondence Learning](https://paperswithcode.co/paper/2512.19687) — [3 paper-native evaluations](https://paperswithcode.co/paper/2512.19687#results).

The PE-AV Large (30 fps, zero-shot) results currently rank first on [Kinetics-400 Zero-Shot Action Classification](https://paperswithcode.co/benchmark/kinetics-400-zero-shot-action-classification?task=video-classification&eval=16898) and [Kinetics-600 Zero-Shot Action Classification](https://paperswithcode.co/benchmark/kinetics-600-zero-shot-action-classification?task=video-classification&eval=16899).

The PE-Core G results currently rank first on [Food-101](https://paperswithcode.co/benchmark/food-101?task=image-classification&eval=7793) and [ImageNet-R](https://paperswithcode.co/benchmark/imagenet-r?task=image-classification&eval=7791).

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?

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

If you'd like to showcase the results in your repository README, you can copy these live leaderboard badges (or use the “Copy PwC badge” button in the Results section):

[](https://paperswithcode.co/api/v1/papers/2504.13180/leaderboard-badge-link?eval=7839)
[](https://paperswithcode.co/api/v1/papers/2504.13181/leaderboard-badge-link?eval=7793)
[](https://paperswithcode.co/api/v1/papers/2504.13181/leaderboard-badge-link?eval=7791)
[](https://paperswithcode.co/api/v1/papers/2512.19687/leaderboard-badge-link?eval=16898)
[](https://paperswithcode.co/api/v1/papers/2512.19687/leaderboard-badge-link?eval=16899)
[](https://paperswithcode.co/api/v1/papers/2512.19687/leaderboard-badge-link?eval=16900)

Kind regards,

Niels

Start with the three linked Papers with Code paper pages and their benchmark result pages. Compare the listed scores, model names, protocols, and openness metadata with the referenced papers, and report any discrepancies or confirm the entries when all details match.

Written by the indexing model from the issue text.