Yolo8 polygon annotations "ignoring corrupt image/label: non-normalized or out of bounds coordinates"

Community Help
1

Hi all,

I am training a yolo v8 object detection model. My annotations are polygons (not bounding boxes). I get plenty of the following warnings when attempting training:
“val: WARNING :warning: C:\image.jpg: ignoring corrupt image/label: non-normalized or out of bounds coordinates [ 1.0064]”

In other posts I have seen that yolo v8 should be able to train based on polygon annotations. I have tried to deactivate “auto-orient” in the preprocessing steps as suggested elsewhere, the problem however persists.

Any help would be appreciated.

Thanks and best regards.

2

Hi @Manuel-Weber-ETH

Is this issue occurring when you are attempting to train an object detection or instance segmentation model using polygon annotations?

Also, could you share the Univese link (or workspace and project ID) of the project that this is occurring with, along with the image file name that’s experiencing this error?

3

Hi @leo

Thank you for your answer. This issue is occurring when I am attempting to train an object detection model using polygon annotations.

It is the following project:
@misc{ spatially-exlicit-stocking-rates_dataset,
title = { Spatially Exlicit Stocking Rates Dataset },
type = { Open Source Dataset },
author = { Carrying Capacities },
howpublished = { \url{ https://universe.roboflow.com/carrying-capacities/spatially-exlicit-stocking-rates } },
url = { https://universe.roboflow.com/carrying-capacities/spatially-exlicit-stocking-rates },
journal = { Roboflow Universe },
publisher = { Roboflow },
year = { 2024 },
month = { jan },
note = { visited on 2024-01-24 },
}

Some of the images that cause the issue (however there are plenty):
RCNX1591_JPGG.rf.b63dabfa3be42015f76daba2b7cdf3de.jpg
RCNX1593_JPGG.rf.19f8f4f6d15ea511cb6007862a066f40.jpg
RCNX1594_JPGG.rf.c67c52fc16a412262bc281b70f705daf.jpg
RCNX1601_JPGG.rf.056d26ac04b4ff99726fc6bee511a669.jpg
RCNX1601_JPGG.rf.0b7b60842a2b2222433f4905964571d9.jpg
RCNX1820_JPGG.rf.4b7382eed3a771f6e2860a4e0b83b51e.jpg
RCNX0406_JPG.rrf.83f879ac84896fd93c97d49e3c0d1105.jpg

Thanks for your help!

4

We are also experiencing the same problem with an instance segmentation model using polygon annotations when importing a dataset to a YOLOv8 format.
All the polygons points are within the image but for some reason when imported some of them aren’t normalized properly which result in the out of bounds error message.

1 Like
5

Hi @Manuel-Weber-ETH @amirt

This issue may be related to a bug we recently fixed. Please generate a new version and follow up if the issue persists.

1 Like
7

This topic was automatically closed 7 days after the last reply. New replies are no longer allowed.