I always like to try these zero-shot models on things outside of the "normal" COCO classes. Here are some chess board queries: Counting: https://imgur.com/KTuQ

I always like to try these zero-shot models on things outside of the "normal" COCO classes. Here are some chess board queries: Counting: https://imgur.com/KTuQ1Bv https://imgur.com/KTuQ1Bv Parse the chess board: https://imgur.com/2zYFK1P https://imgur.com/2zYFK1P (Result): https://imgur.com/Ei4MAl7 https://imgur.com/Ei4MAl7 Few-Shot Object Detection (Pascal VOC): https://imgur.com/gZkDMn8 https://imgur.com/gZkDMn8 Few-Shot Object Detection (simplified): https://imgur.com/Hk8QGMd https://imgur.com/Hk8QGMd Not quite there yet. I've been more impressed with the other new zero-shot multimodal models like Grounding DINO and Azure Dense Captioning. Really looking forward to putting multimodal GPT-4 through its paces as well.