LociArt proposes GRNStyle for controllable local video stylization
Original title: LociArt: A Unified Paradigm for Free-Style Local Video Stylization
The method is trained on 100K synthetic pairs and evaluated on 80 videos with a regional style metric.
Researchers introduce LociArt, a paradigm for free-style local video stylization that specifies where to stylize and which style to apply while leaving other content unchanged. GRNStyle uses a polar style-aware mask and representation-geometry conditioning on Generative Refinement Networks. An automated pipeline of global stylization, object removal, matting, and compositing produced the LociArt-100K dataset. On a benchmark of 80 source videos and 84 masks, the authors report better regional style quality and spatial control than existing methods, with strong temporal consistency.
- LociArt-100K contains 100K local stylization pairs from an automated pipeline.
- The benchmark covers 168 foreground and background stylization cases.
- Style Prototype Similarity measures regional style quality.
- The model can stylize multiple regions despite single-region training pairs.
From the original article
1 min read at huggingface.co
Video stylization typically transforms entire scenes, while local artistic editing requires precise region--style association, spatial control, and temporal consistency. We introduce LociArt, a unified paradigm for free-style local video stylization that explicitly specifies where to stylize and what style to apply while preserving non-target content.
By Yuxuan Chen, Feng Xie, Jiagao Hu · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.