AI Paper Atlas

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

CVPR 2026 · Video › Video-language understanding · approach: Long video understanding · 5 citations (OpenAlex)

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 changedWhereMeasuredEffect
add NTP pre-training (ablation)AutoGaze · AutoGaze recon loss 0.7gazing ratio: 0.209 -> 0.094Improved
add RL post-training (ablation)AutoGaze · AutoGaze recon loss 0.7gazing ratio: 0.102 -> 0.094Improved
add both stages vs none (ablation)AutoGaze · AutoGaze recon loss 0.7gazing ratio: 0.263 -> 0.094Improved
add multi-scale gazing (ablation)AutoGaze · AutoGaze recon loss 0.7gazing ratio: 0.220 -> 0.094Improved
add multi-scale gazing (ablation)AutoGaze · AutoGaze recon loss 0.7latency: 0.467s -> 0.193sImproved
predict 10 tokens at once instead of 1 (ablation)AutoGaze · AutoGaze recon loss 0.7latency: 0.949s -> 0.193sImproved
predict 10 tokens at once instead of 1 (ablation)AutoGaze · AutoGaze recon loss 0.7gazing ratio: 0.074 -> 0.094Hurt
predict 20 tokens instead of 10 (ablation)AutoGaze · AutoGaze recon loss 0.7gazing ratio: 0.094 -> 0.109Hurt
add AutoGaze to NVILA-8B-Video and scale to 1K frames 4KNVILA-8B-Video · HLVidaccuracy (%): 42.5 -> 52.6Improved
add AutoGaze to NVILA-8B-Video and scale to 1K frames 4KNVILA-8B-Video · VideoMME (w/o sub)accuracy (%): 64.2 -> 67.0Improved
prune patches before ViT with AutoGazeMLLM with 128 frames · ViT latencylatency: 2.20s -> 0.55sImproved
use AutoGaze instead of no token reductionMLLM · VideoMME (w/o sub) at 6.25% tokensaccuracy (%): 53.4 -> 52.3No 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.

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

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy