Hello,
we noticed today that after exporting segmentation mask from a semantic segmentation project the masks have the wrong resolution and the annotations inside seems cropped.
Our original dataset had 1920x1080 images and the masks are exported as 1080x1920 with the content half cropped.
No preprocessing/augmentation was used in creating the dataset.
Can you help us? Thanks!
Hi @Dario_Lucantoni is there EXIF data in the images? Did you import the images with EXIF data intact? Additionally, did you apply Auto-Orient as a preprocessing step?
Additionally, if you still find issues, if you can add a bug report here, we can look into this further:
_classes.csv file, annotation file(s), and example images and masks that exported without errors along with images and masks that exported with errors.Hello,
i did an export with Auto-orient but the issue persists. EXIF data on the imported images is the following:
|create|2022-09-28T22:15:21+00:00|
|---|---|
|modify|2022-09-28T22:15:21+00:00|
|cHRM |chunk was found (see Chromaticity, above)|
|gAMA |gamma=0.45454544 (See Gamma, above)|
|IHDR.bit_depth |8|
|IHDR.color_type |2 (Truecolor)|
|IHDR.interlace_method|0 (Not interlaced)|
|IHDR.width,height |1920, 1080|
|sRGB |intent=0 (See Rendering intent)|
I think the issue is related to this other post: Images' mask are in original size after pre-processing when the project is exported as segmentation masks - #2 by Mohamed
How should i proceed?
Thanks.
Ah, no! Sorry to see the issue persists. Yes, if you can file a bug report at the page linked above, and also include the details about the EXIF data from your last reply.
I really appreciate it! And my apologies for the bug!
Also having the same problem👆
We just got an update pushed that should have this fully resolved.