Research map / Video / Video-language understanding
TRACE: Trajectory-Aware Conceps for Explainable Video Understanding
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's abstract only. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes TRACE, a trajectory-aware concept-based explanation method for video understanding models that identifies which source of motion supports a prediction.
Problem
Existing video explainability methods are source-ambiguous (region-centric) or source-incomplete (concept-based with single-object motion vocabularies) when explaining which motion source drives a video model's prediction.
Approach
Frames motion-source under-specification as the key gap and proposes trajectory-aware concepts so explanations capture which motion source (not just whether motion matters) supports a prediction, beyond single-object motion vocabularies.
Limitations
- Abstract is truncated; full text unavailable, so details of method and results are unknown.
Closest papers by meaning
- Disentangled Concepts Speak Louder Than Words: Explainable Video Action Recognition (NeurIPS 2025 · similarity 0.90)
- Motion-Grounded Video Reasoning: Understanding and Perceiving Motion at Pixel Level (CVPR 2025 · similarity 0.88)
- HuMoCon: Concept Discovery for Human Motion Understanding (CVPR 2025 · similarity 0.88)
- Towards Understanding Camera Motions in Any Video (NeurIPS 2025 · similarity 0.87)
- DisMo: Disentangled Motion Representations for Open-World Motion Transfer (NeurIPS 2025 · similarity 0.87)
- Video-of-Thought: Step-by-Step Video Reasoning from Perception to Cognition (ICML 2024 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.