Migrating the EfficientSAM model to the OpenCV model zoo

[Zhang-Yang-Sustech]

EfficientSAM Model Demonstration

I have successfully written the code for the EfficientSAM model and created an interactive demo that enables normal model inference and user interaction. Users can easily use this demo via the command line with the following command:

python demo.py --input /path/to/your/image.jpg

By simply providing the path to the image they wish to segment, the demo will display the image. Upon clicking the object they want to segment in the image, a new window will show the segmentation result, making it user-friendly and straightforward.

Example

Please note that the current model inference speed takes approximately 2 seconds. The camera demo is not yet completed, and I plan to optimize it and add camera functionality in the future.

Performance Testing

The model's benchmark testing has not been conducted yet. I will perform the tests in the future and provide the relevant performance data.

---

Thank you for your attention. I will continue to update and optimize the EfficientSAM model and its demonstration.

[fengyuentau]

Thank you for contribution! Please have a look at my comments below.

[fengyuentau]

The original repo offers EfficientSAM-S and -Ti. Which one is used here? Add a note to describe including version (shasum, md5sum, etc.). Also describe how you convert the model to ONNX (a script would be nice).

Also need to describe how many clicks are required.

[WanliZhong]

Good job👍! Thanks for your contribution. You can check the comment I left. BTW, the benchmark and quantized model part is expected if possible.

[Zhang-Yang-Sustech]

now this pr is only about model and demo, no benchmark

[fengyuentau]

Please fix the following issue and we are going to merge this PR for 4.10.0 release.

[fengyuentau]

You can drop changes in file for now as it affects the benchmark.

[WanliZhong]

LGTM!