4DCodeBench: Benchmarking Agents on Inverse Graphics of Dynamic Scenes
The benchmark asks agents to rebuild moving scenes from video as executable graphics code.
4DCodeBench tests 4D inverse graphics, where a model must infer both scene structure and dynamics. The tasks include real-world videos and synthetic scenes covering deformation, fluid flow, and fracture. The authors report that frontier models with strong static reconstruction still struggle to reproduce complex motion reliably. ArXiv · AI/CL/LG's note
4DCodeBench tests 4D inverse graphics, where a model must infer both scene structure and dynamics. The tasks include real-world videos and synthetic scenes covering deformation, fluid flow, and fracture. The authors report that frontier models with strong static reconstruction still struggle to reproduce complex motion reliably. ArXiv · AI/CL/LG's note
score 5