search

NVIDIA / pix2pixHD

Synthesizing and manipulating 2048x1024 images with conditional GANs
https://tcwang0509.github.io/pix2pixHD/
Other
6.65k stars 1.39k forks source link

question on training on my own dataset. #108

Open imnaren142 opened 5 years ago

imnaren142 commented 5 years ago

Hi,Can you please tell me if I can use this code to generate the human images in color with masks?

bandhanpreeto commented 5 years ago

Which folder did you put ur images in?? I am trying to use RGB image to train my dataset or to even test with my own images....

imnaren142 commented 5 years ago

I have trained the images by putting color images in train_img and label maps in train_label folders.

imnaren142 commented 5 years ago

@tcwang0509 Hi, I have trained the network on my own dataset where RGB image is human color image and label is the mask of the human.The question I have is if I use the trained model on a random mask which is not used in the training and test it with the trained model, is it going to generate new fake human model or existing one?

bandhanpreeto commented 5 years ago

@naren142 how did you convert your images into label maps? I was trying to train with only RGB image without label maps, and it doesn't seem to work. I'm going to use sloth and see if the label maps do work. And I think it will test and create the synthesized image using the trained objects that exist in the program. Let me know what program u used.

imnaren142 commented 5 years ago

I have used PGN network https://github.com/Engineering-Course/CIHP_PGN for mask

shuntos commented 3 years ago

@naren142 Did you generate human body using mask ?