Xiaomi’s MiLM Plus Releases PROVE: Perception-Aligned Object Removal Metrics RC-S and RC-T With a Real-World Video Benchmark
Object removal models have improved faster than the metrics used to judge them. Diffusion erasers now reconstruct shadows, reflections and occluded structure convincingly, yet PSNR, SSIM, LPIPS, ReMOVE and CFD frequently rank their outputs the wrong way. The root cause is structural: erasure is an ill-posed, one-to-many task, so no single ground truth exists to compare against. A team from MiLM Plus, Xiaomi Inc. has released PROVE (Perceptual RemOVal cohErence), accepted at ACM MM 2026, to clos
Get smart on it
Object removal in videos has advanced faster than the metrics used to evaluate it, leaving existing evaluation methods like PSNR and SSIM frequently ranking results incorrectly. A team has released PROVE, a new evaluation system consisting of two perception-aligned metrics that score edited regions locally using deep learning features without requiring a reference video. RC-S measures spatial coherence while RC-T measures temporal consistency, and both significantly outperform existing metrics when compared against human rankings. The release includes code, benchmark videos with varying difficulty levels, and demonstrates practical deployment capability on standard hardware for applications ranging from smartphone photo editing to film post-production.
Related news
State of Open Models: Summer 2026 Observations
As open-source AI models proliferate, tracking their capabilities against closed alternatives helps developers choose which tools suit their needs and budgets.
What We Learned by Reproducing 2,200 papers from ICML
Researchers found significant gaps when attempting to verify published machine learning results, raising questions about reproducibility standards in AI research.
