Open-Vocabulary Attribute Detection

Skibidi vibes meet sigma grindset in the realm of open-vocabulary attribute detection. This paper dives deep into the dank world of vision-language models, where the toilet of traditional object detection gets flushed away. With a clean test set covering 117 attributes across 80 MS COCO classes, researchers are lowkey redefining how we probe object attributes. Positive and negative annotations? That is goon-level detail! They showcase a baseline method that ships the future of open-vocabulary tasks. Get ready for a wild ride as models flex their attribute detection skills and push the edge of what is possible in computer vision!