search

NVlabs / SPADE

Semantic Image Synthesis with SPADE
https://nvlabs.github.io/SPADE/
Other
7.59k stars 982 forks source link

Generate RGB Images #92

Open SaidaSaad opened 4 years ago

SaidaSaad commented 4 years ago

Hello I would like to ask if SPADE takes fixed dimension maps because I am trying to use your pre trained model for COCO Stuff dataset to generate RGB Images for my own Sementic maps but i got that error We got that error complaining about the dimensions which really different form the input that SPADE use RuntimeError: Given groups=1, weight of size [1024, 184, 3, 3], expected input[1, 186, 8, 8] to have 184 channels, but got 186 channels instead.

The dimension of the semantic maps that i am using : 256*256. I would like to know the properties of the semantic maps that SPADE used to generate RGB Images. and if you can recommend me any way to solve this problem.

Thanks

Kevin-X-Wang commented 4 years ago

Have you found anything out about this? I'm currently having this issue as well.

segalinc commented 4 years ago

Same problem when using custom dataset and trying to use --instance_dir. my images are 1920x1080

SaidaSaad commented 4 years ago

@Kevin-X-Wang not yet

segalinc commented 4 years ago

have you tried adding label_nc ... for me it worked adding this parameter. the testing configuration is not optimal as you still have to pass label_dir image_dir label_nc no_instance params. basically the same one you pass when training.

wanganyi1102 commented 1 year ago

I changed the semantic map from rgb to greyscale and it worked.