Research map / Video / Video-language understanding
Time Blindness: Why Video-Language Models Can't See What Humans Can?
Ujjwal Upadhyay, Mukul Ranjan, Zhiqiang Shen, Mohamed Elhoseiny
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 chain-of-thought prompting (ablation) | Video-VLMs · SpookyBench | accuracy: 0% -> 0% | No clear effect |
| raise frame rate from 1 to 30 FPS (ablation) | Video-VLMs (4 models) · SpookyBench | accuracy: 0.0 at all FPS | No clear effect |
| raise frame rate from 1 to 30 FPS (ablation) | humans · SpookyBench | accuracy: 0.0 -> 95.6 | Improved |
| finetune on 400 SpookyBench videos (ablation) | InternVL2.5-8B and Qwen2-VL-7B · SpookyBench test | accuracy: 0% -> 0% | No clear effect |
| increase model scale (2B to 78B) and use video-specialized models (ablation) | Video-VLMs · SpookyBench | accuracy: 0% across 2B-78B | No clear effect |
| overlay pre-computed motion boundaries on noisy frames (ablation) | Qwen2-VL-7B · SpookyBench | accuracy overall: 0.0% -> 51.54% | Improved |
| overlay pre-computed motion boundaries on noisy frames (ablation) | GPT-4o · SpookyBench | accuracy overall: 0.0% -> 59.10% | Improved |
| overlay motion boundaries on dynamic-scene videos (ablation) | GPT-4o and Qwen2-VL-7B · SpookyBench Videos | accuracy: 0.0% -> 3.51% (GPT-4o), 1.75% (Qwen2-VL-7B) | Mixed |
| finetune frame-level video classifiers to detect foreground noise (ablation) | VJEPA-2 and DINOv3 · foreground noise detection | accuracy: ~50% (chance) | No clear effect |
Closest papers by meaning
- PerceptionLM: Open-Access Data and Models for Detailed Visual Understanding (NeurIPS 2025 · similarity 0.88)
- MotionBench: Benchmarking and Improving Fine-grained Video Motion Understanding for Vision Language Models (CVPR 2025 · similarity 0.87)
- VideoHallu: Evaluating and Mitigating Multi-modal Hallucinations on Synthetic Video Understanding (NeurIPS 2025 · similarity 0.87)
- MMBench-Video: A Long-Form Multi-Shot Benchmark for Holistic Video Understanding (NeurIPS 2024 · similarity 0.87)
- VidHalluc: Evaluating Temporal Hallucinations in Multimodal Large Language Models for Video Understanding (CVPR 2025 · similarity 0.87)
- Re-thinking Temporal Search for Long-Form Video Understanding (CVPR 2025 · 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.
