Building a Precise Video Language with Human-AI Oversight

A festival programme printed on cool paper: Garamond headlines, typewriter tags, and a black screening room just past the title.

linzhiqiu.github.io/papers/chaiBuilding a Precise Video Language with Human-AI Oversight
Screenshot of https://linzhiqiu.github.io/papers/chai/, © the authors. Shown for reference. Request removal
Venue
CVPR 2026
Year
2026
First author
Zhiqiu Lin
Field
video-language data
Template
custom single-file (inline CSS with custom properties)
Theme
Light

Checked against the rendered page. The dark Ours-vs-baselines grid is a single video in the hook band just below the first viewport (starts at y=846), not in the 1440x900 first screen. Instrument Serif is declared but unused; mono weight 700 renders with the 600 face.

Why it is here

The most deliberate page in the pool. A serif display headline sits over a mono badge row, the copy is short, and the dark comparison grid of Ours against baselines gives the first screen a real visual payoff. Numbered overview cards and the illustrated problem/solution strip keep the rhythm going all the way down.

Overview

Color Palette

Typography

EB Garamond
Building a Precise Video Language with Human-AI Oversight Abstract · Method · Results · BibTeX — 0123456789
DM Sans
Building a Precise Video Language with Human-AI Oversight Abstract · Method · Results · BibTeX — 0123456789
Roboto Mono
Building a Precise Video Language with Human-AI Oversight Abstract · Method · Results · BibTeX — 0123456789

Spacing & Shape

Page Modules

Guidelines

More like this