Open fjremnav opened 3 years ago
Does this work for you? https://github.com/NVIDIA/semantic-segmentation/tree/sdcnet#pre-trained-models Look at the pre-trained models for sdccityscapes and FlowNet2_
@fitsumreda
My understanding of sdcnet usage is to generate additional annotations and augment to cityscape datasheet. Is this correct? Does this pretrained model "pretrained_models/cityscapes_best.pth[1071MB, WideResNet38 backbone]" use scdnet augmentation in the training?
Thanks,
Yes, it was trained with sdcnet augmentation.
@fitsumreda
Thanks for the answer. I am looking for a way, like using optical flow to improve temporal consistency of segmentation results for a video. Do you know anyone who has worked on this area?
Thanks,
Is there a demo.py for sdcnet? Where can I download pretrained sdcnet?
Thanks,