Grounded-Segment-Anything VS GroundingDINO

Grounded SAM is a project that tries to combine these steps in a single workflow, but I am not sure how far it has come along. Might be worth checking out, but it also isn't too difficult to combine the three models by hand.

We build a simple demo ImageBind-SAM here which aims to segment with different modalities

GroundingDINO install failed. Please submit an issue to https://github.com/IDEA-Research/Grounded-Segment-Anything/issues.

This is because C++ is somehow not compiled. Check https://github.com/IDEA-Research/Grounded-Segment-Anything/issues/53 too see whether it’s working, otherwise search through similar issues. Let me know which solution works. Remember that ‘export’ on windows is ‘set’, and you should make sure that CUDA_HOME exists in your environment variable.

I'm glad to do so after I implement this.

here is the GitHub link: https://github.com/IDEA-Research/Grounded-Segment-Anything

Some of the foundation/base models include: * GroundedSAM (Segment Anything Model) * DETIC * GroundingDINO

While Segment Anything can detect objects based on text prompts, that's not its strong suite. To get best results, folks usually combine it with Grounding DINO, which is a great object detection model. You run Grounding DINO with text prompt "skirt", this gives you a bounding box that you pass to Segment Anything, which gives you a segmentation mask that you can then use for inpainting with SD.

Now, let's delve into the technological aspects of this project. The combination of Facebook's Segment Anything and Grounding Dino tools will automate annotations for image processing, which is key to this AI endeavor. I'm also intrigued by Mojo, a new programming language designed specifically for AI developers, which will soon be open-source.

Right now we are trying out grouding dino on this but it is giving a lot of noise and detecting things that are not cracks.