Research map / Video / Video-language understanding
Violin: A Large-Scale Dataset for Video-and-Language Inference
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 subtitles to statement input (ablation) | Violin baseline, BERT · Violin test set | accuracy: 54.20 -> 66.05 | Helped |
| Add full video with image features to statement and subtitles (ablation) | Violin baseline, BERT · Violin test set | accuracy: 66.05 -> 67.60 | Helped |
| Use full video rather than one middle frame (ablation) | Stmt+Subtt+Vis with Img+BERT · Violin test set | accuracy: 66.60 -> 67.60 | Helped |
| Use BERT instead of GloVe (ablation) | Violin baseline · Violin test set | accuracy: 60.33 -> 67.60 | Helped |
| Use detection features instead of image features (ablation) | Stmt+Subtt+Vis with BERT · Violin test set | accuracy: 67.60 -> 67.84 | Helped |
| Use C3D features instead of image features (ablation) | Stmt+Subtt+Vis with BERT · Violin test set | accuracy: 67.60 -> 67.23 | Hurt |
| Use LXMERT joint multimodal features instead of the best detection-feature baseline (ablation) | Violin baseline · Violin test set | accuracy: 67.84 -> 66.25 | Hurt |
| Use adversarially matched rather than human-written negative statements (ablation) | Stmt+Subtt+Vis, Image+BERT · Violin test set | accuracy: 61.99 -> 67.60 | Helped |
| Add video features (ablation) | Stmt+Subtt, BERT, on conversation reasoning and human dynamics · Violin reasoning categories | accuracy | No clear effect |
| Add video features (ablation) | Stmt+Subtt, BERT, on visual recognition and action recognition · Violin reasoning categories | accuracy: 67.19 -> 67.97; 70.75 -> 73.58 | Helped |
Closest papers by meaning
- LongVideoBench: A Benchmark for Long-context Interleaved Video-Language Understanding (NeurIPS 2024 · similarity 0.89)
- Two Causally Related Needles in a Video Haystack (NeurIPS 2025 · 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)
- InternVid: A Large-scale Video-Text Dataset for Multimodal Understanding and Generation (ICLR 2024 · similarity 0.87)
- SF2T: Self-supervised Fragment Finetuning of Video-LLMs for Fine-Grained Understanding (CVPR 2025 · similarity 0.87)
- Unleashing Hour-Scale Video Training for Long Video-Language Understanding (NeurIPS 2025 · 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.