I am attempting to tag images with CVAT to generate Segmentation Class images to convert to polygons. When I export the dataset as Segmentation Mask 1.1, the images in the SegmenationClass directory are pixelated or have jagged edges on any lines that have an angle.
This is the original image that I am tagging with "M" tag applied.
In the above image you can see that there are 13 points defined when drawing the polygon. In the below image you can see a close up of the polygon line.
After exporting the dataset with the Segmentation Mask 1.1 format, I get the following.
Image exported from CVAT with Segmentation Mask 1.1
And here is a close up of the at image showing pixelation.
Closeup of pixelation of exported image
When I use python script to convert the output image to a polygon, it creates all the pixelation and jagged edges as points. My end goal is to use these polygons as a training set for yolov8. When I use the predication generated from this training, it shows heavy pixelation on the identification of the segment in the prediction.
- Is this expected behavior of CVAT and the export of dataset with format Segmentation Mask 1.1?
- Is there a way/configuration of CVAT to have higher resolution images with no pixelation/jagged edges?
- How can I get clean images that will generate limited number of points in polygon when ran through python script?