keejkrej

I have found a way to embed the template into the raw image, use its bounding box to detect similar objects on the embedded image, then use one of the...

> you mind sharing how you did this approach? I'm curious given we may have a similar use case You could check out [segment_with_template_embedded.py](https://github.com/keejkrej/sam3/blob/main/mytest/segment_with_template_embedded.py) in my fork of sam3. It...

> > Hi sam3 team, > > thanks for your awesome work! Is it possible to transfer the embedding learned from the bounding box of an object in one image...

> Sorry for taking up your time—I’ve found a flaw in the cross-image solution. > > When I was testing it for convenience, I used the horizontally flipped prompt image...

I have succeeded with cross image segmentation.

For the cats, I used another image as template. @echochoc have you checked that your bounding box yields good results for the first image, otherwise it will surely struggle with...

[cross_image_predictor.py](https://github.com/keejkrej/sam3/blob/14b55465244e51002660178e1f89283211ff81a3/scripts/cross_image_predictor.py) [sam3_image.py](https://github.com/keejkrej/sam3/blob/14b55465244e51002660178e1f89283211ff81a3/sam3/model/sam3_image.py) [sam3_image_processor.py](https://github.com/keejkrej/sam3/blob/14b55465244e51002660178e1f89283211ff81a3/sam3/model/sam3_image_processor.py) [Commit](https://github.com/keejkrej/sam3/commit/14b55465244e51002660178e1f89283211ff81a3)

> > [cross_image_predictor.py](https://github.com/keejkrej/sam3/blob/14b55465244e51002660178e1f89283211ff81a3/scripts/cross_image_predictor.py) [sam3_image.py](https://github.com/keejkrej/sam3/blob/14b55465244e51002660178e1f89283211ff81a3/sam3/model/sam3_image.py) [sam3_image_processor.py](https://github.com/keejkrej/sam3/blob/14b55465244e51002660178e1f89283211ff81a3/sam3/model/sam3_image_processor.py) [Commit](https://github.com/keejkrej/sam3/commit/14b55465244e51002660178e1f89283211ff81a3) > > I tried your code and observed a very interesting thing: If the sample image contains only a single subject (a cat), the...

This time I did not duplicate the cat, just extended the image by adding more background, you can see that the model is so stubborn to predict an object at...

> > This time I did not duplicate the cat, just extended the image by adding more background, you can see that the model is so stubborn to predict an...