Researchers have introduced TRACE, a new benchmark designed to evaluate object understanding capabilities in computer vision models after irreversible physical damage, such as that caused by fire. The benchmark includes over 21,000 synthetic scenes with paired pristine and degraded object progressions. To address the significant performance degradation observed in existing models like RF-DETR and InternVL3.5, the paper proposes a Feature Recovery Module (FRM). This plug-and-play module maps degraded features to pristine-aligned representations, showing an average performance improvement of over 12% across various vision-language tasks, with greater gains on more severely damaged objects. AI
IMPACT This research could improve AI's ability to analyze damaged environments, aiding in disaster response and reconstruction efforts.
RANK_REASON The cluster describes a new academic paper introducing a benchmark and a module for a specific computer vision task.
Read on Hugging Face Daily Papers →
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →
