Research map / Video / Video-language understanding
AdsQA: Towards Advertisement Video Understanding
Xinwei Long, Kai Tian, Peng Xu, Guoli Jia, Jingxuan Li, Sa Yang, Yihua Shao, Kaiyan Zhang, Che Jiang, Hao Xu, Yang Liu, Jiaheng Ma and 1 more
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 |
|---|---|---|---|
| train on uncurated data instead of curated 500 QA (ablation) | ReAd-R (Qwen2-VL-7B) · AdsQA | strict accuracy: 18.6 -> 13.2 | Hurt |
| use curated data instead of uncurated data (ablation) | ReAd-R (Qwen2-VL-7B) · AdsQA | relaxed accuracy: 36.2 -> 44.9 | Improved |
| use graded reward (0.5 partial) instead of strict reward (ablation) | ReAd-R (Qwen2-VL-7B) · AdsQA | strict accuracy: 15.1 -> 18.6 | Improved |
| add prompt constraints on think/answer tags (ablation) | ReAd-R (Qwen2-VL-7B) · AdsQA | strict accuracy: 16.8 -> 18.6 | Improved |
| apply RL (GRPO) instead of SFT on same 500 QA | Qwen2-VL-7B · AdsQA | strict accuracy: 14.1 -> 18.6 | Improved |
| SFT on 1,053 uncurated videos instead of limited data (ablation) | Qwen2-VL-7B (SFT) · AdsQA | strict accuracy: 14.1 -> 16.0 | Improved |
| increase VOT chain length from 3 to 5 (ablation) | VOT (Qwen2-VL-7B) · AdsQA | strict accuracy: 17.0 -> 13.6 | Hurt |
| add MCTSr search to Qwen2-VL-7B | Qwen2-VL-7B · AdsQA | relaxed accuracy: 41.0 -> 43.4 | Improved |
| add ReAd-R RL fine-tuning to Qwen2-VL-7B | Qwen2-VL-7B · AdsQA | strict accuracy: 17.2 -> 18.6 | Improved |
| apply ReAd-R to Qwen2.5-VL-7B | Qwen2.5-VL-7B · AdsQA | strict accuracy: 23.0 -> 25.0 | Improved |
Closest papers by meaning
- VideoQA in the Era of LLMs: An Empirical Study (IJCV 2025 · similarity 0.89)
- VideoAds for Fast-Paced Video Understanding (ICCV 2025 · similarity 0.88)
- CG-Bench: Clue-grounded Question Answering Benchmark for Long Video Understanding (ICLR 2025 · similarity 0.86)
- Can Video LLMs Refuse to Answer? Alignment for Answerability in Video Large Language Models (ICLR 2025 · similarity 0.86)
- VideoTree: Adaptive Tree-based Video Representation for LLM Reasoning on Long Videos (CVPR 2025 · similarity 0.86)
- MMR-V: What's Left Unsaid? A Benchmark for Multimodal Deep Reasoning in Videos (ICLR 2026 · 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.
