Research map / Video / Video-language understanding
AXG-Reasoner: Error Detection and Explanation in Long Task Videos with Vision-Language Models
Shih-Po Lee, Ehsan Elhamifar
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add AXG subaction decomposition and keyframes to Qwen2.5-VL | Qwen2.5-VL-32B · EgoPER error explanation (GT segments) | explanation similarity score (%): 3.6 -> 17.4 | Improved |
| add AXG to InternVL3.5 | InternVL3.5-14B · CaptainCook4D error explanation (GT segments) | explanation similarity score (%): 18.0 -> 21.2 | Improved |
| use VideoTree keyframe selection instead of uniform sampling | Qwen2.5-VL-32B · EgoPER error explanation (GT segments) | explanation similarity score (%): 3.6 -> 5.0 | No clear effect |
| add AXG with GTG2Vid TAS to Qwen2.5-VL for detection | Qwen2.5-VL-32B · EgoPER error detection F1@0.5 | F1@0.5: 20.4 -> 33.4 | Improved |
| add AXG with GTG2Vid TAS to Qwen2.5-VL for detection | Qwen2.5-VL-32B · CaptainCook4D error detection F1@0.5 | F1@0.5: 15.3 -> 28.5 | Improved |
| add AXG on GT action segments | Qwen2.5-VL-32B · EgoPER error detection (GT segments) | F1: 56.7 -> 63.6 | Improved |
| add AXG on GT action segments | Qwen2.5-VL-32B · EgoPER normal-segment F1 (GT segments) | F1 on normal segments (N.): 87.1 -> 80.1 | Hurt |
| increase number of subaction clusters K from 5 to 20 (ablation) | AXG · EgoPER and CaptainCook4D tasks | error explanation score: Burritos S 27.0 -> 30.0; Tea S 17.0 -> 19.0; Oatmeal S 20.0 -> 22.0 | Mixed |
| increase number of subaction clusters K from 5 to 20 (ablation) | AXG · EgoPER Oatmeal (GT segments) | error detection F1: 65.1 -> 61.3 | Hurt |
| use single-subaction AXG for temporal subaction segmentation | InternVL3.5-14B · EgoPER actions with one subaction | explanation similarity score (%): 18.4 -> 25.0 | Improved |
| InternVL3.5 instead of Qwen2.5-VL as the VLM for explanation | AXG · EgoPER error explanation | explanation similarity score (%): 17.4 -> 24.6 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- VideoTree — VideoTree: Adaptive Tree-based Video Representation for LLM Reasoning on Long Videos (CVPR 2025) · better on 3 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Look, Remember and Reason: Grounded Reasoning in Videos with Language Models (ICLR 2024 · similarity 0.88)
- Black Swan: Abductive and Defeasible Video Reasoning in Unpredictable Events (CVPR 2025 · similarity 0.87)
- CollabVR: Collaborative Video Reasoning with Vision-Language and Video Generation Models (NeurIPS 2026 · similarity 0.87)
- VERA: Explainable Video Anomaly Detection via Verbalized Learning of Vision-Language Models (CVPR 2025 · similarity 0.87)
- Act2See: Emergent Active Visual Perception for Video Reasoning (CVPR 2026 · similarity 0.87)
- MMR-V: What's Left Unsaid? A Benchmark for Multimodal Deep Reasoning in Videos (ICLR 2026 · similarity 0.86)
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.
