Introducing UniRef++: Advanced Object Segmentation in Spatial and Temporal Domains
Key Features:
Unified Model: UniRef++ seamlessly handles segmentation tasks:
Referring Image Segmentation (RIS)
Few-Shot Segmentation (FSS)
Referring Video Object Segmentation (RVOS)
Video Object Segmentation (VOS)
Core Component: UniFusion module
Integrates reference information efficiently
Utilizes flash attention for high efficiency
Compatibility: Acts as a plug-in for foundational models like SAM
UniRef++ is the official extended implementation from ICCV 2023's UniRef.
Stay tuned for more updates!
Code: https://github.com/FoundationVision/UniRef
Paper: [Paper link]
https://t.me/deeplearning_ai
