segment anything #2711
-
|
Would you expect the output of the segment anything model (SAM) using the VisualPrompter to match the output from the original FAIR version? I am not getting the same thing. |
Beta Was this translation helpful? Give feedback.
Replies: 2 comments 3 replies
-
|
Never mind. I was loading the image myself, instead of using kornia.io.load_image, so it was not scaled to between 0 and 1. Perhaps a warning could be printed if the image is not in the expected range? At the very least, it would be nice if the documentation/tutorial mentioned this. |
Beta Was this translation helpful? Give feedback.
-
|
I think it makes sense to have a warning if it's not in range and comment on the documentation. Could you contribute a PR with these changes? |
Beta Was this translation helpful? Give feedback.
-
|
Sure. Which is preferred, the ImagePrompter or the VisualPrompter |
Beta Was this translation helpful? Give feedback.
-
|
ImagePrompter is an alias that should be removed sooner or later. I’m more in favour of updating properly documentation rather than runtime checking which means adding extra O(n) tensor traverse. General rule in Kornia is operating floating point tensors in range 0/1 |
Beta Was this translation helpful? Give feedback.
-
|
Just circling back. I noticed that this is already mentioned in the documentation for VisualPrompter .set_image. I have created a PR to update the kornia documentation: #2738 |
Beta Was this translation helpful? Give feedback.
Never mind. I was loading the image myself, instead of using kornia.io.load_image, so it was not scaled to between 0 and 1. Perhaps a warning could be printed if the image is not in the expected range? At the very least, it would be nice if the documentation/tutorial mentioned this.