Attend Before Attention: Efficient and Scalable Video Understanding via Autoregressive Gazing

A linen-bound field journal: light Figtree on parchment, hairline-underlined names and brass-edged notes.

autogaze.github.ioAttend Before Attention: Efficient and Scalable Video Understanding via Autoregressive Gazing
Screenshot of https://autogaze.github.io/, © the authors. Shown for reference. Request removal
Venue
CVPR 2026
Year
2026
First author
Baifeng Shi
Field
video understanding
Template
Clarity template (Shikun Liu), customized
Theme
Light

Checked against the rendered page. v1 values corrected from the render: body is 20px/300 (not 18px), chips 13px/500 (not 11.7px), title 37px/500.

Why it is here

Restrained warm-cream gradient, a light title weight, small-caps link chips and a narrow text column. It is a quiet, minimal style with good typographic restraint. The video still shows a loading spinner, which is a minor capture flaw.

Overview

Color Palette

Typography

Figtree
Attend Before Attention: Efficient and Scalable Video Understanding vi Abstract · Method · Results · BibTeX — 0123456789
Fira Code
Attend Before Attention: Efficient and Scalable Video Understanding vi Abstract · Method · Results · BibTeX — 0123456789

Spacing & Shape

Page Modules

Guidelines

More like this