Research map / Video / Video-language understanding
Visual Semantic Role Labeling for Video Understanding
Arka Sadhu, Tanmay Gupta, Mark Yatskar, Ram Nevatia, Aniruddha Kembhavi
Check 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 |
|---|---|---|---|
| Use SlowFast rather than I3D for verb prediction | VidSitu verb classification · VidSitu Test | Rec@5 | Improved |
| Pretrain on Kinetics (ablation) | VidSitu verb classifiers · VidSitu | Rec@5: ≈10 to 16 points | Improved |
| Add a Fast pathway to Slow+NL (ablation) | Slow+NL verb classifier · VidSitu Test | Acc@1: 29.03 -> 46.37 | Improved |
| Add Non-Local blocks to I3D without Kinetics pretraining (ablation) | I3D verb classifier · VidSitu Test | Acc@1: 31.91 -> 31.43 | Hurt |
| Add Non-Local blocks to Kinetics-pretrained I3D (ablation) | Kinetics-pretrained I3D verb classifier · VidSitu Test | Acc@1: 29.87 -> 38.42 | Improved |
| Condition role prediction on video features (ablation) | Semantic role prediction · VidSitu Test | CIDEr: 36.48 -> 44.95 | Improved |
| Contextualize events with a transformer encoder (ablation) | Vid TxDec role predictor · VidSitu Test | CIDEr: 44.95 -> 47.25 | Improved |
| Use I3D rather than SlowFast visual features for role prediction (ablation) | Vid TxEncDec · VidSitu Test | CIDEr: 47.25 -> 48.51 | Improved |
| Add verb descriptions to video-only relation prediction (ablation) | VidSitu event relation classifier · VidSitu Test | Macro-Acc: 30.97 -> 32.89 | Improved |
| Add argument descriptions after including verb descriptions (ablation) | VidSitu event relation classifier · VidSitu Test | Macro-Acc: 32.89 -> 32.98 | No clear effect |
Closest papers by meaning
- Grounded Video Situation Recognition (NeurIPS 2022 · similarity 0.92)
- OmniViD: A Generative Framework for Universal Video Understanding (CVPR 2024 · similarity 0.89)
- VELOCITI: Benchmarking Video-Language Compositional Reasoning with Strict Entailment (CVPR 2025 · similarity 0.88)
- E.T. Bench: Towards Open-Ended Event-Level Video-Language Understanding (NeurIPS 2024 · similarity 0.88)
- SF2T: Self-supervised Fragment Finetuning of Video-LLMs for Fine-Grained Understanding (CVPR 2025 · similarity 0.88)
- HENASY: Learning to Assemble Scene-Entities for Interpretable Egocentric Video-Language Model (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.
