NVIDIA / pix2pixHD

Synthesizing and manipulating 2048x1024 images with conditional GANs
https://tcwang0509.github.io/pix2pixHD/
Other
6.67k stars 1.39k forks source link

How to train pix2pixHD on mask to image? #212

Open HankYangg opened 4 years ago

HankYangg commented 4 years ago

My data is pairs of mask and x-ray image. Both are 1024*1024. I want to synthesize more images. But, the results in the training process look very blurry. This is my command.

python train.py --name xray --batchSize 4 --gpu_ids 0 --label_nc 0 --no_instance --dataroot ./datasets/xray

Can somebody give me some hints how to train?

JiaojiaoYe1994 commented 4 years ago

I would guess the result would be very blurry, since one big step pf original work is to combine instance map as auxiliary information to augment boundary information. And for your task from mask to image, it would be even harder for Net to learn information.

tzl1 commented 4 years ago

I have the same problem with you. How did you solve it?