FIX: region outflow #204

[glenntfung]

If the user annotates a region outside of the image, the downloaded data will only reflect portions within the image.

[vercel]

@glenntfung is attempting to deploy a commit to the Suman Kunwar's projects Team on Vercel.

A member of the Team first needs to authorize it.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Preview	Comments	Updated (UTC)
annotate-lab	Ready	Preview	Comment	Aug 31, 2025 11:12pm

[sumn2u]

Thanks for this PR. I was able to annotate beyond the image area and download the annotated coordinates. I think we need to prevent users from annotating outside the image on the client side.

The annotated co-ordinates.

1 0.763911 0.145161 0.301210 0.148387
1 1.000000 0.154032 0.000000 0.308065

[glenntfung]

Thanks for this PR. I was able to annotate beyond the image area and download the annotated coordinates. I think we need to prevent users from annotating outside the image on the client side.

The annotated co-ordinates.

1 0.763911 0.145161 0.301210 0.148387
1 1.000000 0.154032 0.000000 0.308065

Sure. Yes, my clamping logic only applies when the user intends to annotate something inside the image but accidentally drags the mouse outside. I will fix the client as well.

[glenntfung]

Added a new commit; this should prevent the annotator from working beyond the image :)

[sumn2u]

Added a new commit; this should prevent the annotator from working beyond the image :)

This is great, thanks. There seems to be a regression: the Masked Image and Annotated Image download functionality is not working.

[glenntfung]

Added a new commit; this should prevent the annotator from working beyond the image :)

This is great, thanks. There seems to be a regression: the Masked Image and Annotated Image download functionality is not working.

Thanks for pointing out. I reverted the server change. With the updated client we don't need to do any data manipulation. This recovered Masked Image downloading. Annotated Image download does not work in the current master branch as I tested just now :(

[glenntfung]

Added a new commit; this should prevent the annotator from working beyond the image :)

This is great, thanks. There seems to be a regression: the Masked Image and Annotated Image download functionality is not working.

Thanks for pointing out. I reverted the server change. With the updated client we don't need to do any data manipulation. This recovered Masked Image downloading. Annotated Image download does not work in the current master branch as I tested just now :(

Can you confirm this @sumn2u? I think we need another issue/PR for Annotated Image download if this is also the case on your machine as well, to keep things clean. If this is not the case, then it's my problem and I will try to find it out. Thanks!

[sumn2u]

Added a new commit; this should prevent the annotator from working beyond the image :)

This is great, thanks. There seems to be a regression: the Masked Image and Annotated Image download functionality is not working.

Thanks for pointing out. I reverted the server change. With the updated client we don't need to do any data manipulation. This recovered Masked Image downloading. Annotated Image download does not work in the current master branch as I tested just now :(

Can you confirm this @sumn2u? I think we need another issue/PR for Annotated Image download if this is also the case on your machine as well, to keep things clean. If this is not the case, then it's my problem and I will try to find it out. Thanks!

The annotated image is working fine on my machine with recent changes. Did you first save the image and checked the box before downloading it?

[glenntfung]

I think things are fine now. The problem I had, but not you, should be due to the input image mode. I added a convert to the server so we might not have a problem. My test image was the A Lab's logo in the repo.

[sumn2u]

I think things are fine now. The problem I had, but not you, should be due to the input image mode. I added a convert to the server so we might not have a problem. My test image was the A Lab's logo in the repo.

What do you think about checking the image mode first and then converting if necessary? This might prevent unnecessary conversions.

image = Image.open(BytesIO(response.content))
if image.mode != "RGBA":
   image = image.convert("RGBA")

[glenntfung]

I think things are fine now. The problem I had, but not you, should be due to the input image mode. I added a convert to the server so we might not have a problem. My test image was the A Lab's logo in the repo.

What do you think about checking the image mode first and then converting if necessary? This might prevent unnecessary conversions.

image = Image.open(BytesIO(response.content))
if image.mode != "RGBA":
   image = image.convert("RGBA")

Yes! That's better. I did this.

[sumn2u]

LGTM

[sumn2u]

Fixes #204

[sumn2u]

Great work @glenntfung.