Research map / Video / Video-language understanding
CACR: Reinforcing Temporal Answer Grounding in Instructional Video via Candidate-Aware Causal Reasoning
Muge Qi, Rong Fu, Pengbin Feng, Xianda Li, Yu Cai, Yifu Guo, Shizhe Zhang, Simon James Fong, Lei Ma, Bin Li
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 VBCS Top-K candidates (CACR-TopK vs TimeZero) (ablation) | TimeZero · MedVidQA | mIoU: 31.34 -> 59.04 | Improved |
| add VBCS Top-K candidates (CACR-TopK vs TimeZero) (ablation) | TimeZero · VehicleVQA | mIoU: 42.41 -> 62.88 | Improved |
| add VBCS Top-K candidates (CACR-TopK vs TimeZero) (ablation) | TimeZero · CMIVQA | mIoU: 29.22 -> 39.01 | Improved |
| add subtitle summary and pre-answer (ablation) | TimeZero · MedVidQA | mIoU: 31.34 -> 35.42 | Improved |
| add subtitle summary alone (ablation) | TimeZero · CMIVQA | mIoU: 29.22 -> 27.85 | Hurt |
| add pre-answer alone (ablation) | TimeZero · CMIVQA | mIoU: 29.22 -> 27.12 | Hurt |
| add pre-answer to subtitle summary on top of Top-K (Group4 vs CACR-Cap) (ablation) | CACR · VehicleVQA | mIoU: 65.39 -> 72.11 | Improved |
| add frame description to Top-K + pre-answer + subtitle (Group3 vs Group4) (ablation) | CACR · MedVidQA | mIoU: 59.65 -> 55.63 | Hurt |
| add frame description to Top-K + pre-answer + subtitle (Group3 vs Group4) (ablation) | CACR · VehicleVQA | mIoU: 72.11 -> 69.04 | Hurt |
| add frame description to Top-K + pre-answer + subtitle (Group3 vs Group4) (ablation) | CACR · CMIVQA | mIoU: 45.37 -> 44.60 | Hurt |
| increase number of candidates K from 1 to 5, then 5 to 9 (cumulative max IoU) | VBCS · VehicleVQA | mIoU: 43.82 -> 49.07 | Improved |
| replace Top-K only with Top-K + Pre-Answer (ablation) | CACR · VehicleVQA | mIoU: 62.88 -> 70.85 | Improved |
Closest papers by meaning
- TimeCraft: Navigate Weakly-Supervised Temporal Grounded Video Question Answering via Bi-directional Reasoning (ECCV 2024 · similarity 0.90)
- LLCP: Learning Latent Causal Processes for Reasoning-based Video Question Answer (ICLR 2024 · similarity 0.90)
- Temporal Reasoning Transfer from Text to Video (ICLR 2025 · similarity 0.88)
- STAIR: Spatial-Temporal Reasoning with Auditable Intermediate Results for Video Question Answering (AAAI 2024 · similarity 0.88)
- Not Another Text Benchmark: Putting the “Visual" Back in Visual Question Answering for Large Video Models (NeurIPS 2026 · similarity 0.88)
- MoReVQA: Exploring Modular Reasoning Models for Video Question Answering (CVPR 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.
