search

NVIDIA / vid2vid

Pytorch implementation of our method for high-resolution (e.g. 2048x1024) photorealistic video-to-video translation.
Other
8.57k stars 1.2k forks source link

How to train Mask R-CNN for cityscapes? #103

Open TuanHAnhVN opened 5 years ago

TuanHAnhVN commented 5 years ago

as @forfun1012 metioned https://github.com/NVIDIA/vid2vid/issues/61, we use Mask R-CNN to get instance maps. But @tcwang0509 said "This is actually not very important, as the network wasn't trained on Cityscapes and doesn't perform really well." So I want to train Mask R-CNN for cityscapes again to get results better. But I don't have instance's dataset (growth truth)

Help me by guiding me how to generate the above dataset!

Thank you!

TillBeemelmanns commented 5 years ago
  1. Register & Download the cityscapes dataset from https://www.cityscapes-dataset.com/
  2. Use the cityscapes to coco converter: https://github.com/facebookresearch/Detectron/blob/master/tools/convert_cityscapes_to_coco.py You can choose the classes that which should be considered in the instance generation.
  3. Start from a pretrained Mask R-CNN network https://github.com/matterport/Mask_RCNN#installation and finetune on cityscapes in coco format