---
format: "aidr-story-markdown/v1"
id: "1c233fb3a02c31f99dceb2e66562fa765ed1b28fced8908ab44bb02f92177c44"
canonical_url: "https://aidr.today/1c233fb3?lang=en"
title: "PhyProbe: Rethinking Physical Consistency Evaluation in Generated Videos"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-10-14T00:00:00.000Z"
category: "Research"
topics: ["reasoning"]
source_urls: ["https://huggingface.co/papers/2609.38377"]
summary: "PhyProbe is a new evaluator that extracts features from a frozen pretrained spatio-temporal encoder and maps them to a scalar physical consistency violation score using a lightweight scoring head. It is trained with a unified objective combining pairwise ranking, regression on noisy scalar annotations, and anchor-based calibration over heterogeneous supervision sources. The approach aims to address limitations of existing methods that rely on off-the-shelf vision-language models or fine-tuned evaluators trained on human annotations."
---

# PhyProbe: Rethinking Physical Consistency Evaluation in Generated Videos

> [Open the canonical story](<https://aidr.today/1c233fb3?lang=en>)

**Published:** 2026-10-14T00:00:00.000Z
**Category:** Research
**Topics:** reasoning

## Summary

PhyProbe is a new evaluator that extracts features from a frozen pretrained spatio\-temporal encoder and maps them to a scalar physical consistency violation score using a lightweight scoring head\. It is trained with a unified objective combining pairwise ranking, regression on noisy scalar annotations, and anchor\-based calibration over heterogeneous supervision sources\. The approach aims to address limitations of existing methods that rely on off\-the\-shelf vision\-language models or fine\-tuned evaluators trained on human annotations\.

## Sources

- [Story source](<https://huggingface.co/papers/2609.38377>)