This is a repository for a collection of technical demos made by Annotation-AI.
Every demo doesn't use GPU resources, and they are served on CPU only instances.
One of the most appealing applications of Segment Anything Model (SAM) is Everything, which detects and segments all objects in an image.
However, it requires 1,024 inferences per a single image, and due to the reason, it is hard to be feasible for service providers who don’t have massive GPU resources.
In order to handle this issue, we re-implemented the Everything algorithm in iterative manner that is better for CPU only environments.
Mostly, we can expect comparable results to the original Everything within 1/5 number of inferences (e.g. 1024 vs 200), and it takes under 10 seconds to search for masks on a CPU upgrade instance
(8 vCPU, 32GB RAM) of Huggingface space.
We named it Fast Segment Everything
.
You can run the DEMO in HuggingFace Space.
Based on Fast Segment Everything
algorithm, we built a number of applications for promptable segmentation.: text, image, drawing, and click.
Title | GIF | Demo |
---|---|---|
Segment everything with a text prompt | HuggingFace Space | |
Segment everything with an image prompt | HuggingFace Space | |
Segment everything with a drawing prompt | HuggingFace Space | |
Segment similar things! (Single click) | HuggingFace Space |
Person re-identification(Re-ID) aims at retrieving a person of interest across multiple non-overlapping cameras. We implement Re-ID applications for various situations.
Title | GIF | Demo |
---|---|---|
Re-ID with query & gallery images | HuggingFace Space |