[Project Page] [arXiv] [PDF] [Slides] [BibTeX]

• We introduce PosSAM, an open-vocabulary panoptic segmentation model that generates class and instance-aware masks with excellent generalization to a variety of visual concepts by unifying SAM and CLIP in an end-to-end trainable framework.
• We develop a novel Local Discriminative Pooling (LDP) module to enhance discriminative CLIP features with class-agnostic SAM features for an unbiased OV classification.
• We introduce the Mask-Aware Selective Ensembling algorithm to adaptively discern between seen and unseen classes by leveraging IoU and LDP confidence scores for each image.
• We conduct extensive experiments and demonstrate superior performance over existing state-of-the-art open-vocabulary panoptic segmentation methods across multiple benchmark datasets.

'Coming Soon...!!!'

'Coming Soon...!!!'