TRACE benchmark tackles object understanding after irreversible fire damage with feature recovery

UIUC-CS · hf · 2026-09-15

UIUC released TRACE, a benchmark for post-fire object understanding. It introduces a feature recovery module to restore degraded representations caused by irreversible fire damage, improving detection and vision-language performance under severe physical damage.

The work targets an understudied setting: how well models understand and recover information about objects whose appearance has been largely destroyed.

Original post →

More from Research

Research channel →