Get unusal yaml file and wrong label text files

Hello, I found some problems while exporting my project to yolov8.

First. yaml file result is change:
The existing yaml file result is:
train: …/train/images
test: …/test/images

nc: 1
names: [‘Target2’]
but now it changed like this:
path: …/datasets/roboflow

train: train/images
val: valid/images
test: test/images

0: Target2

Second, There are changes to the label text files generated.

In the case of the existing label text file, if it is object detection, 5 numbers, including the class and 4 coordinates, are expressed in the file. If it is segmentation, the number as many as the number of classes and polygon coordinates is expressed in the file. Currently, it is expressed as a fixed class and 8 numbers.

This problem has been occurring since January 23rd, and other accounts are also experiencing the same problem.
However, we confirmed that the version created before that was exported in a normal file format.

How can I solve this? Any suggestions are welcome. Thank You!

hey, did it mess up when you read the data? I’m facing the same issue and it messed up my object detection annotate

I think the problem occurs in the process of generating a new version within the project or exporting it as a zip file.
As mentioned in the above article, when the version generated before January 23 is exported, normal annotations are extracted.

Hi @kael

There are different variations of the YOLOv8 YAML format that are all valid and there can be changes in how we create export files from time to time. From what I can see, the YAML format change doesn’t seem to be the cause of your issues.

You are correct in that segmentation annotation files should include the class ID and polygon coordinates.

Could you share your project’s Universe link (or workspace and project ID) so we can look into this for you?

@leo hi,

sorry to jump to discuss, on my workspace got a certain error, the first one is what happens when kael says, and the second one is when I’m exporting (had 18 classes) it results in classes 8-18 for example not getting read. it was readable with the previous version, for now, it’s not. when I was processing the data and it messed up my object classification result.

Hi @dimasilham_7

To make sure I understand you right, you’re experiencing two issues:

  1. The issue kael brought up of having only eight coordinates on the annotation file
    • Like I replied to them, could you please share your project’s Universe link?
  2. Classes 8-18 “not being read”
    • Could you explain what you mean when you say they aren’t getting “read”? What package, website or app is “reading” it? Are there error messages? If so, please send that as well.
  1. here’s my universe link you can check on it, and yes the YAML file format change started on the 23 of January like what kael experienced it.
  2. what I mean by read is, when I’m visualization my class with Python, it just can read only 8 classes from 22 classes for this example on image, it seems normal if i just produce 8 classes only but not if over that 8

Hey @kael and @dimasilham_7

This issue (@dimasilham_7 - your first issue, at least) seems related to a bug we recently fixed. Please generate a new version and follow up if the issue persists.

hello @leo
It works well without errors in my project. Thank you for fixing the bug and replying my topic!!

A post was split to a new topic: Issues with classes not exporting

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