Stubbi
Rankings
Benchmark

4DCodeBench

Test whether coding agents can reconstruct dynamic scenes from video.

Checked Oct 21, 20264DCodeBench authors

Published leaderboard

Explore scores and evaluation settings on the publisher’s leaderboard.

Open leaderboard

Coverage

200 tasks: 100 real videos and 100 simulated videos.

Methodology

Agents write executable graphics programs. Compare appearance, geometry and motion using five metric families; report model-judge and human pairwise preferences separately.

People & organizations

9

Sources

Original paperHacker News discussion