Closed islamtashfiq closed 7 years ago
The link for the GTA source dilation model has now been added.
Does the model trained on ImageNet before training on the GTA5 data?
Yes, this model has been pretrained on ImageNet (challenge data).
Just to be clear, did you use this dilation10_cityscapes_deploy.prototxt for training?
during training and evaluation the prototxt is programmatically generated. See train.py in dilation codebase for an example.
Can you share the weights of dilation network that was trained on GTA5 data and/or cityscapes?