search

Wizaron / instance-segmentation-pytorch

Semantic Instance Segmentation with a Discriminative Loss Function in PyTorch
https://arxiv.org/abs/1708.02551
GNU General Public License v3.0
458 stars 95 forks source link

Cityscapes instance #5

Closed zhengduoru closed 6 years ago

zhengduoru commented 6 years ago

hello, me again. Execuse me. I read the paper _Semantic Instance Segmentation with a Discriminative Loss Function It is amazing that your model can be used with cityscape dataset. I wanna know what should I do to apply your model on cityscape dataset.

Wizaron commented 6 years ago

Hey.

Cityscapes dataset has multiple classes, whereas CVPPP has only two classes (foreground (leaf) and background). Hence, this code currently assumes that there are two classes.

arch.py, model.py and prediction.py should be edited in order to allow multiple classes.

I am going to make an update to support multiple classes soon.

zhengduoru commented 6 years ago

Looking forward to multiple classes~ Thank you.

Wizaron commented 6 years ago

I have just created a branch named "cityscapes". You can find codes to train on cityscapes dataset in there. However, network architecture (number of filters etc.) needs to be updated according to the dataset.

I will merge it to master, after some tests and hyper-parameter tuning.

pvthuy commented 4 years ago

Sorry, any updates? @Wizaron I have tried with the branch "cityscapes" for multi-class instance segemntation, but I can't train the model using your code.