Hello everyone. Can you tell me (in particular @dusty_nv ) with which version of the Cityscapes dataset the FNC-ResNet18 network was trained for semantic segmentation in the nano jetson? On the official page of the dataset there are different versions (ex: gtCourse, gtFine, leftImg8bit etc). Which of this was chosen? Thank you.


Based on the output image, it should be trained with fine annotations.
Please see the comment below:


Hi @forflafor, those FCN-ResNet18 models were trained with the Cityscapes gtCourse annotations. I recall trying both course/fine, and the course annotations performed better.

1 Like

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