Segment unknown scene with unsupervised learning using UNet diffusion features with DiffCut
Segment unknown scene with unsupervised learning using UNet diffusion features with DiffCut
Zero-Shot Image Segmentation via Recursive Normalized Cut on Diffusion Features
arXiv paper abstract https://arxiv.org/abs/2406.02842
arXiv PDF paper https://arxiv.org/pdf/2406.02842
GitHub https://github.com/PaulCouairon/DiffCut
Project page https://diffcut-segmentation.github.io
Foundation models have emerged as powerful tools across various domains including language, vision, and multimodal tasks.
While prior works have addressed unsupervised image segmentation, they significantly lag behind supervised models.
… use a diffusion UNet encoder as … vision encoder and introduce DiffCut, an unsupervised zero-shot segmentation method that solely harnesses the output features from the final self-attention block.
… demonstrate that the utilization of these diffusion features in a graph based segmentation algorithm, significantly outperforms … state-of-the-art methods on zero-shot segmentation.
… leverage a recursive Normalized Cut algorithm that softly regulates the granularity of detected objects and produces … segmentation maps that … capture intricate image details.
… work highlights the remarkably accurate semantic knowledge embedded within diffusion UNet encoders that could then serve as foundation vision encoders for downstream tasks …
Stay up to date. Subscribe to my posts https://morrislee1234.wixsite.com/website/contact
Web site with my other posts by category https://morrislee1234.wixsite.com/website