Attend Before Attention: Efficient and Scalable Video Understanding via Autoregressive Gazing
2026-03-12 • Computer Vision and Pattern Recognition
Computer Vision and Pattern Recognition
AI summaryⓘ
The authors created AutoGaze, a smart tool that helps large language models understand long and detailed videos better by ignoring repetitive parts. Instead of looking at every single pixel, AutoGaze picks only the important patches needed to reconstruct the video with little loss of information. This approach makes video processing much faster and allows analysis of very long, high-quality videos. They also introduced a new test with 4K videos to show that using AutoGaze improves performance on video question answering tasks.
multi-modal large language modelsvision transformersspatiotemporal redundancyautoregressive selectionnext-token predictionreinforcement learningvideo benchmarks4K-resolution videovideo question answeringlong-form video understanding
Authors
Baifeng Shi, Stephanie Fu, Long Lian, Hanrong Ye, David Eigen, Aaron Reite, Boyi Li, Jan Kautz, Song Han, David M. Chan, Pavlo Molchanov, Trevor Darrell, Hongxu Yin
Abstract
Multi-modal large language models (MLLMs) have advanced general-purpose video understanding but struggle with long, high-resolution videos -- they process every pixel equally in their vision transformers (ViTs) or LLMs despite significant spatiotemporal redundancy. We introduce AutoGaze, a lightweight module that removes redundant patches before processed by a ViT or an MLLM. Trained with next-token prediction and reinforcement learning, AutoGaze autoregressively selects a minimal set of multi-scale patches that can reconstruct the video within a user-specified error threshold, eliminating redundancy while preserving information. Empirically, AutoGaze reduces visual tokens by 4x-100x and accelerates ViTs and MLLMs by up to 19x, enabling scaling MLLMs to 1K-frame 4K-resolution videos and achieving superior results on video benchmarks (e.g., 67.0% on VideoMME). Furthermore, we introduce HLVid: the first high-resolution, long-form video QA benchmark with 5-minute 4K-resolution videos, where an MLLM scaled with AutoGaze improves over the baseline by 10.1% and outperforms the previous best MLLM by 4.5%. Project page: https://autogaze.github.io/.