Open Vocabulary Panoptic Segmentation With Retrieval Augmentation
The episode discusses Nafis Sadeq, Qingfeng Liu, and Mostafa El-Khamy's paper on Open Vocabulary Panoptic Segmentation with Retrieval Augmentation. The hosts explain how retrieval augmentation helps overcome domain shift issues when using vision-language models like CLIP for identifying unseen objects. They detail the pipeline involving Grounding DINO and SAM for mask generation, showing significant performance improvements.