How to run on cityscpaes dataset

facebookresearch / maskrcnn-benchmark

Fast, modular reference implementation of Instance Segmentation and Object Detection algorithms in PyTorch.

MIT License

9.29k stars 2.5k forks source link

How to run on cityscpaes dataset #279

Open ywlng opened 5 years ago

ywlng commented 5 years ago

❓ Questions and Help

I use the ./configs/cityscapes/e2e_mask_rcnn_R_50_FPN_1x_cocostyle.yaml config file to run the demo. But the output shows no predictions. The heatmap is as follow: 000000 How can I fix this problem?

fmassa commented 5 years ago

We currently do not provide pre-trained models for cityscapes dataset, which would explain why you have no predictions, as you are using a model from scratch.

ywlng commented 5 years ago

I found that pre-trained model are from a PR (https://github.com/facebookresearch/maskrcnn-benchmark/pull/232), I will ask the author for more details.

fmassa commented 5 years ago

There are no pre-trained models for this dataset, they are all initialized from ImageNet.

ywlng commented 5 years ago

ok, thanks