Image to Image translation like pix2pix

[iamrishab]

Paired image to image translation tasks as in pix2pixHD (domain A -> domain B with paired images of the same name), Over in the SPADE repo the same question was asked multiple times, however, with no answer yet:

NVlabs/SPADE#46
NVlabs/SPADE#112

In this PR there are some changes in the dataloader and some changes in the network architecture parameters as well.

python -u train.py --name test --dataset_mode pix2pix --dataroot datasets/dataset_wall_reports_v0/ --no_instance --label_nc 0 --batchSize 1

[AugustLee93]

Paired image to image translation tasks as in pix2pixHD (domain A -> domain B with paired images of the same name), Over in the SPADE repo the same question was asked multiple times, however, with no answer yet:

NVlabs/SPADE#46 NVlabs/SPADE#112

In this PR there are some changes in the dataloader and some changes in the network architecture parameters as well.

python -u train.py --name test --dataset_mode pix2pix --dataroot datasets/dataset_wall_reports_v0/ --no_instance --label_nc 0 --batchSize 1

@iamrishab Thanks your work!! I tried it, it can works, but when I do the testing , it's weired that it always went to generate the train_A folder, what is the standard test command ? Thanks

[iamrishab]

@AugustLee93
Updated the data loader for test. Folder structure should be same as in pix2pixHD

python test.py --name test/ --dataset_mode pix2pix --dataroot datasets/test/ --gpu_ids 0,1,2,3 --which_epoch latest --no_instance --label_nc 0 --batchSize 4

[NuclearKAI]

useful

[khanhdc]

What a good PR. it goes smoothly.
I have a question about the output of training, my input images are 1024x1024 however, during the training process, the generated synthesized_image over the training epochs and on the testing (in results folder) show the smaller image size (320x320)