Research map / Video / Video-language understanding
VRR-QA: Visual Relational Reasoning in Videos Beyond Explicit Cues
Sirnam Swetha, Rohit Gupta, Parth Parag Kulkarni, David G Shatwell, Jeffrey A Chan Santiago, Nyle Siddiqui, Joseph Fioresi, Mubarak Shah
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 a structured spatial and narrative reasoning prompt (ablation) | GPT-4.1 Full · VRR-QA | accuracy (avg, %): 54.30 -> 58.20 | Improved |
| add a structured spatial and narrative reasoning prompt (ablation) | GPT-4.1 Mini · VRR-QA | accuracy (avg, %): 40.30 -> 45.07 | Improved |
| increase frames from 8 to 16 (ablation) | video LLMs · VRR-QA | accuracy | Improved |
| increase frames from 16 to 32 (ablation) | video LLMs · VRR-QA | accuracy: LLaVA-NeXT-Video 33.9 -> 32.56 | No clear effect |
| increase frames from 4 to 32 for inferred counting (ablation) | Qwen2.5-VL-32B · VRR-QA Inferred Counting | accuracy (%): 27.91 -> 41.86 | Improved |
| increase frames from 1 to 32 (ablation) | Qwen2.5-VL-7B · VRR-QA | accuracy (avg, %): 38.18 -> 45.07 | Improved |
| scale Qwen2.5-VL from 3B to 32B (ablation) | Qwen2.5-VL · VRR-QA | accuracy: approximately +2% accuracy | Improved |
| scale GPT-4.1 from Mini to Full (ablation) | GPT-4.1 · VRR-QA | accuracy (avg, %): 40.30 -> 54.30 | Improved |
| use reasoning model GPT O3 instead of GPT 4.1 | GPT family · VRR-QA | accuracy (avg, %): 54.3 -> 64.1 | Improved |
| enable thinking tokens (ablation) | Claude 4.5 Sonnet · VRR-QA | accuracy (avg, %): 42.8 -> 48.6 | Improved |
| enable thinking tokens (ablation) | GPT 5.2 · VRR-QA | accuracy (avg, %): 50.8 -> 56.8 | Improved |
| allow annotators to view the video multiple times for counting (ablation) | non-expert human baseline · VRR-QA Inferred Counting | accuracy (%): 65.9 -> 82.8 | Improved |
Closest papers by meaning
- MovieRecapsQA: A Multimodal Open-Ended Video Question-Answering Benchmark (CVPR 2026 · similarity 0.91)
- VideoEspresso: A Large-Scale Chain-of-Thought Dataset for Fine-Grained Video Reasoning via Core Frame Selection (CVPR 2025 · similarity 0.90)
- MMR-V: What's Left Unsaid? A Benchmark for Multimodal Deep Reasoning in Videos (ICLR 2026 · similarity 0.90)
- LongVideoBench: A Benchmark for Long-context Interleaved Video-Language Understanding (NeurIPS 2024 · similarity 0.90)
- VTQA: Visual Text Question Answering via Entity Alignment and Cross-Media Reasoning (CVPR 2024 · similarity 0.90)
- Not Another Text Benchmark: Putting the “Visual" Back in Visual Question Answering for Large Video Models (NeurIPS 2026 · similarity 0.89)
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.
