Research map / Video / Video-language understanding
CaST-Bench: Benchmarking Causal Chain-Grounded Spatio-Temporal Reasoning for Video Question Answering
Mingfang Zhang, Jingjing Pan, Ashutosh Kumar, Rajat Saini, Mustafa Erdogan, Hsuan-Kung Yang, Caixin Kang, Yifei Huang, Yoichi Sato, Quan Kong
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 self-generated causal chain before answering (ablation) | Gemini-2.5-Pro · CaST-Bench | MCQ accuracy: 45.98 -> 50.34 | Improved |
| add self-generated causal chain before answering (ablation) | GPT-5 · CaST-Bench | MCQ accuracy: 43.90 -> 46.32 | Improved |
| add self-generated causal chain before answering (ablation) | GLM-4.1V-9B · CaST-Bench | MCQ accuracy: 38.09 -> 39.55 | Improved |
| add self-generated causal chain before answering (ablation) | Qwen2.5-VL-7B · CaST-Bench | MCQ accuracy: 40.90 -> 41.09 | No clear effect |
| provide ground-truth causal chain instead of self-generated (ablation) | Gemini-2.5-Pro · CaST-Bench | MCQ accuracy: 50.34 -> 75.61 | Improved |
| provide ground-truth causal chain instead of self-generated (ablation) | Qwen2.5-VL-7B · CaST-Bench | MCQ accuracy: 41.09 -> 59.68 | Improved |
| provide visual video input instead of text-only | Gemini-2.5-Pro · CaST-Bench | MCQ accuracy: 23.14 -> 50.34 | Improved |
| use Thinking variant instead of Instruct | Qwen3-VL-8B · CaST-Bench | MCQ accuracy: 43.13 -> 39.16 | Hurt |
| scale InternVL-3.5 from 8B to 30B-A3B | InternVL-3.5 · CaST-Bench | MCQ accuracy: 40.95 -> 44.53 | Mixed |
Closest papers by meaning
- VideoEspresso: A Large-Scale Chain-of-Thought Dataset for Fine-Grained Video Reasoning via Core Frame Selection (CVPR 2025 · similarity 0.91)
- Two Causally Related Needles in a Video Haystack (NeurIPS 2025 · similarity 0.90)
- Look, Remember and Reason: Grounded Reasoning in Videos with Language Models (ICLR 2024 · similarity 0.90)
- Temporal Chain of Thought: Long-Video Understanding by Thinking in Frames (NeurIPS 2025 · similarity 0.90)
- SOK-Bench: A Situated Video Reasoning Benchmark with Aligned Open-World Knowledge (CVPR 2024 · similarity 0.90)
- ReXTime: A Benchmark Suite for Reasoning-Across-Time in Videos (NeurIPS 2024 · similarity 0.90)
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.
