MiLM Plus releases PROVE, metrics for evaluating object removal models

A team from Xiaomi's MiLM Plus has released PROVE, a new evaluation framework for assessing object removal models. The framework introduces two perception-aligned metrics—RC-S for spatial coherence and RC-T for temporal consistency—along with PROVE-Bench, a two-tier real-world video benchmark. Object removal is an ill-posed task with multiple plausible solutions, making traditional metrics like PSNR and SSIM unsuitable for assessment. PROVE addresses this by using sliding-window Maximum Mean Discrepancy over DINOv2 features to score edited regions locally, without requiring a reference video. The framework is released as an open-source Apache 2.0 PyTorch repository, requiring Python 3.10+, PyTorch 2.6+, Transformers 4.51+, and DINOv2-giant weights.