Brief

Researchers have introduced CRONOS, a new benchmark designed to test the physical consistency of video generation models. This benchmark, built in Unreal Engine, evaluates how well models predict physical events when visual inputs like scene context, viewpoint, and object appearance are altered. Initial evaluations using CRONOS revealed that current open-source video generation models struggle with counterfactual physical consistency, showing degraded performance when conditions change. AI

Researchers have introduced StereoGenBench, a new synthetic dataset created in Unreal Engine for evaluating stereo generation and view synthesis. This benchmark provides meticulously controlled camera baselines, intrinsics, scene depth, and motion data across multiple calibrated views. It aims to enable precise measurement of how sensitive stereo generation models are to varying baseline configurations and ensure consistency across different camera setups. AI