Feature/api sam predict boxes

[Dutch77]

return bounding boxes on dino and sam API

[light-and-ray]

You breakes ui and extensions which use it

[light-and-ray]

You can save it inside image object instead

[Dutch77]

It's appended stuff. Which UI and extension does it break please?

It's really shame it doesn't return this info. Can you suggest any other way than meta data of image? I need to automate some stuff and extracting info from image seems really non-standard way of how to do stuff. (actualy cannot do this without some heavy implementation in lib I use)

[light-and-ray]

You can use your fork in your app, I don't think it's a big problem

Can you suggest any other way than meta data of image?

I mean here:

return Image.fromarray(show_boxes(image_np, boxes_filt_numpy.astype(int), show_index=True)), ..., boxes_filt

turn it into

res = Image.fromarray(show_boxes(image_np, boxes_filt_numpy.astype(int), show_index=True))
res.boxes_filt = boxes_filt
return res, ...

In apy.py get them by

[x.boxes_filt for x in sam_output_mask_gallery[:3]]

and

dino_output_img.boxes_filt

[light-and-ray]

sam_predict and dino_predict are used in extension's UI, and in my extension https://github.com/light-and-ray/sd-webui-replacer