Research map / Video / Video-language understanding
ClearText-Video: A Large-Scale Text-Centric Video Dataset Bridging Video Restoration and Scene-Text Enhancement
JINLONG LI, Jiaming Ding, Dingfu Lu, Malcolm Hsiu, Chuang Ke, Kangning Yang, Bochen Guan, Lan Fu, Jie Cai, Huiming Sun, Zibo Meng
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 |
|---|---|---|---|
| degrade video by blur instead of low resolution (x4 bicubic) (ablation) | 16 MLLMs averaged · CTVid spatial VideoQA | accuracy drop from HQ: -3.14 (LR) vs -6.05 (blur) | Hurt |
| fine-tune Qwen2.5-VL-7B on CTVid training set | Qwen2.5-VL-7B · CTVid spatial VideoQA (HQ) | accuracy: 49.99 -> 58.43 | Improved |
| fine-tune Qwen2.5-VL-7B on CTVid training set | Qwen2.5-VL-7B · CTVid spatial VideoQA (HQ) | overconfidence ratio: 21.36 -> 1.73 | Improved |
| fine-tune Qwen2.5-VL-7B on CTVid training set | Qwen2.5-VL-7B · CTVid spatial VideoQA (HQ) | uncertainty-aware accuracy: 60.66 -> 43.71 | Hurt |
| apply video restoration (DOVE, MIMO, S3Diff) to degraded input before MLLM (ablation) | MLLMs on spatial VideoQA · CTVid spatial VideoQA | accuracy | Mixed |
| restore degraded video before MLLM on temporal QA (ablation) | MLLMs on temporal VideoQA · CTVid temporal VideoQA | accuracy | Mixed |
| switch from perceptual-metric-best SR to text-aware metrics | image SR methods · CTVid image SR | Acc_rec: SwinIR PSNR 28.18, Acc_rec 36.76; S3Diff PSNR 26.05, Acc_rec 38.42 | Mixed |
| replace direct MLLM reasoning with OCR+LLM pipeline (ablation) | spatial VideoQA · CTVid spatial VideoQA | accuracy | Hurt |
| use proprietary instead of open-source MLLMs on temporal QA (non-Claude models) | Gemini, GPT-4o-mini · CTVid temporal VideoQA | accuracy | No clear effect |
Closest papers by meaning
- VCR: A Task for Pixel-Level Complex Reasoning in Vision Language Models via Restoring Occluded Text (ICLR 2025 · similarity 0.89)
- TOPA: Extending Large Language Models for Video Understanding via Text-Only Pre-Alignment (NeurIPS 2024 · similarity 0.88)
- MMOU: A Massive Multi-Task Omni Understanding and Reasoning Benchmark for Long and Complex Real-World Videos (NeurIPS 2026 · similarity 0.87)
- Not Another Text Benchmark: Putting the “Visual" Back in Visual Question Answering for Large Video Models (NeurIPS 2026 · similarity 0.87)
- Panda-70M: Captioning 70M Videos with Multiple Cross-Modality Teachers (CVPR 2024 · similarity 0.87)
- SlowFocus: Enhancing Fine-grained Temporal Understanding in Video LLM (NeurIPS 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.
