Research map / Video / Video-language understanding
Attend Before Attention: Efficient and Scalable Video Understanding via Autoregressive Gazing
Baifeng Shi, Stephanie Fu, Long Lian, Hanrong Ye, David Eigen, Aaron Reite, Jan Kautz, Boyi Li, David M. Chan, Trevor Darrell, Pavlo Molchanov, Hongxu Yin
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 NTP pre-training (ablation) | AutoGaze · AutoGaze recon loss 0.7 | gazing ratio: 0.209 -> 0.094 | Improved |
| add RL post-training (ablation) | AutoGaze · AutoGaze recon loss 0.7 | gazing ratio: 0.102 -> 0.094 | Improved |
| add both stages vs none (ablation) | AutoGaze · AutoGaze recon loss 0.7 | gazing ratio: 0.263 -> 0.094 | Improved |
| add multi-scale gazing (ablation) | AutoGaze · AutoGaze recon loss 0.7 | gazing ratio: 0.220 -> 0.094 | Improved |
| add multi-scale gazing (ablation) | AutoGaze · AutoGaze recon loss 0.7 | latency: 0.467s -> 0.193s | Improved |
| predict 10 tokens at once instead of 1 (ablation) | AutoGaze · AutoGaze recon loss 0.7 | latency: 0.949s -> 0.193s | Improved |
| predict 10 tokens at once instead of 1 (ablation) | AutoGaze · AutoGaze recon loss 0.7 | gazing ratio: 0.074 -> 0.094 | Hurt |
| predict 20 tokens instead of 10 (ablation) | AutoGaze · AutoGaze recon loss 0.7 | gazing ratio: 0.094 -> 0.109 | Hurt |
| add AutoGaze to NVILA-8B-Video and scale to 1K frames 4K | NVILA-8B-Video · HLVid | accuracy (%): 42.5 -> 52.6 | Improved |
| add AutoGaze to NVILA-8B-Video and scale to 1K frames 4K | NVILA-8B-Video · VideoMME (w/o sub) | accuracy (%): 64.2 -> 67.0 | Improved |
| prune patches before ViT with AutoGaze | MLLM with 128 frames · ViT latency | latency: 2.20s -> 0.55s | Improved |
| use AutoGaze instead of no token reduction | MLLM · VideoMME (w/o sub) at 6.25% tokens | accuracy (%): 53.4 -> 52.3 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- VideoChat-Flash — VideoChat-Flash: Hierarchical Compression for Long-Context Video Modeling (ICLR 2026) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- MotionBench: Benchmarking and Improving Fine-grained Video Motion Understanding for Vision Language Models (CVPR 2025 · similarity 0.88)
- Time-R1: Post-Training Large Vision Language Model for Temporal Video Grounding (NeurIPS 2025 · similarity 0.88)
- ReWind: Understanding Long Videos with Instructed Learnable Memory (CVPR 2025 · similarity 0.88)
- StreamGaze: Gaze-Guided Temporal Reasoning and Proactive Understanding in Streaming Videos (NeurIPS 2026 · similarity 0.88)
- BOLT: Boost Large Vision-Language Model Without Training for Long-form Video Understanding (CVPR 2025 · similarity 0.88)
- MVBench: A Comprehensive Multi-modal Video Understanding Benchmark (CVPR 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.
