AI TrackerFrontier desk · AEST
Back to news
normalaitech

4DCodeBench: Benchmarking Agents on Inverse Graphics of Dynamic Scenes

arXiv cs.AI · 3 Oct 2026, 03:58 AEST

4DCodeBench is a new benchmark for 4D inverse graphics that focuses on code generation for reconstructing dynamic scenes from video. It evaluates agents' ability to translate visual observations into executable graphics programs, highlighting challenges in reconstructing complex dynamics.

Key facts

Read aloud

4DCodeBench is a new benchmark for 4D inverse graphics that focuses on code generation for reconstructing dynamic scenes from video. It evaluates agents' ability to translate visual observations into executable graphics programs, highlighting challenges in reconstructing complex dynamics. We introduce 4DCodeBench, a benchmark for 4D inverse graphics through code generation, in which agents reconstruct dynamic scenes from video as executable graphics programs. To evaluate this capability, we curate a set of real-world videos and construct synthetic scenes spanning diverse physical phenomena, including deformation, fluid flow, and fracture. We perform extensive benchmarking of frontier models, finding that strong static reconstruction capabilities do not yet translate into reliable reconstruction of complex dynamics.

Sources in this cluster