Open JackeyGHD1 opened 3 years ago
we have it as part of the code. Please see. https://github.com/NVIDIA/semantic-segmentation/blob/main/train.py#L226
I had the same question, if I understand it correctly your code drops the prediction as well as the prob separately. In your paper you mention a threshold of 0.9 but this is not part of your repository correct?
In the paper , the mentioned "hard labelling strategy", Could you release the code of the hard labelling strategy?