[Python] How can I perform zero-shot inference in RegionCLIP after I extract my concept pool

How can I perform zero-shot inference in RegionCLIP after I extract my concept pool

I have successfully extracted my own concept features using RegionCLIP. However, I'm unsure of the steps required to perform zero-shot inference with these custom features. I have the extracted concept pool saved in a .pth file, but I don’t know how to proceed with integrating this into the inference pipeline. I expected to use my custom concept pool for zero-shot inference, where the model would recognize objects in new images based on the embeddings in my custom concept pool. However, I'm unclear on the exact commands and configurations needed to run the inference correctly.

Continue reading...