All labels have same coordinates for all images

Object detection - I uploaded around 500 images and annotated them.

But in labels folder for all images coordinates are same. label for each images is same except class id. Like in the below both labels 0.5, 0.5, 1, 1 are same.

4 0.5 0.5 1 1
2 0.5 0.5 1 1

Apologies if it’s simple question. I am new to computer vision and training the models. I followed your video on youtube and worked on this. Can you please help me where I am wrong and how can I solve this problem.

Hi @Sukumar_Reddy,

Sorry to hear you’re having issues. I have some questions so I can try to help you better. What do your annotations look like?

Is this happening when you are viewing your downloaded dataset? If so, what type of format did you export with? (ex: COCO, YOLOv8, TFRecord, etc)

Also if you could share what Roboflow video you were following, that might be helpful.

Hi @leo …Thank you for your reply. Yes I downloaded dataset and checked in datasets\train\labels folder. Each file has same coordinates (0.5 0.5 1 1)

I downloaded using Yolov8 by clicking on get snippet and selected zip download option and my model URL is Vehicle detector Object Detection Dataset and Pre-Trained Model by Vehicle Detection

Please let me know if anything needed.

Hi @Sukumar_Reddy

I took a look at your dataset and I think I understand what might’ve gone wrong.

The first image in the test split was the following one, which did have the coordinates you mentioned:

The annotation format for YOLOv8 is class_id center_x center_y width height and since this entire image seems to be one object, that annotation makes sense. (It’s highlighting the entire image)

It appears you used the Isolate Objects preprocessing feature. It’s intended to turn object detection datasets (drawing boxes around items) to a classification dataset (labeling one entire image) which I assume you don’t want.

You can remove this preprocessing step when generating a new version, which should solve your issues.

Hope this helps