Can YOLOv4 Be Trained with Polygon Annotations Instead of Rectangular Bounding Boxes?

Accelerated Computing Intelligent Video Analytics TAO Toolkit
1

Please provide the following information when requesting support.

• Hardware (T4/V100/Xavier/Nano/etc) : A6000
• Network Type (Detectnet_v2/Faster_rcnn/Yolo_v4/LPRnet/Mask_rcnn/Classification/etc) : Yolo_v4
• TLT Version (Please run “tlt info --verbose” and share “docker_tag” here) : 5.3.0
• Training spec file(If have, please share here)
• How to reproduce the issue ? (This is for errors. Please share the command line and the detailed log here.)

I’m working on a project where my objects of interest are better represented by polygons rather than rectangular bounding boxes.
Is it possible to train a YOLOv4 model using polygon annotations instead of rectangles?

2

Looks like a feature request. This is not supported by default. May I know that how does the object look like? Is it a text object?

3

Nope, its not a text object. its an object which is slightly tilted. if not available what is the best option to perform object detection on the object which are tilted.

4

If you are training only one class, you can try to use OCDNet. After training, it can generate polygon inference result.
Or, you can consider to use segmentation network. For example, segformer or Unet.

5

There is no update from you for a period, assuming this is not an issue anymore. Hence we are closing this topic. If need further support, please open a new one. Thanks

6

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.