Feature Recovery for Object Understanding After Irreversible Fire Damage

  • 类型:arxiv
  • 标识:2609.12078
  • 链接:https://arxiv.org/abs/2609.12078
  • 主分类:evaluation
  • 形态:benchmark
  • TLDR:Objects in post-fire environments often undergo irreversible physical transformations that change their geometry, material state, and visual appearance. Detecting and identifying these remnants is critical for locating hazards, reconstructing pre-incident contents, and inventorying losses. Unlike standard image corruptions, these degradations affect the physical structure of the object itself. To study this setting, we introduce TRACE, a transformation-aware benchmark for post-fire object understanding. TRACE contains 21.4K real-image-grounded synthetic scenes and paired object-level pristine-
  • 待LLM分类:否
  • 来源文件
  • /inbox/tom/_candidates/2026-09-15-rag-retrieval-reranking-candidates.json
  • /inbox/tom/_candidates/2026-09-15-agent-rag-longcontext-candidates.json