Colorization problem
songyn95 opened this issue · comments
Hello, I want to convert infrared images at night into RGB. It is required that the effect of RGB should be consistent with that during the day.
- I see that the author has many colorization methods (pix2pix, interactive deep colorization). Which method is the most suitable?
- My goal is to convert infrared images into RGB, so I use RGB images for training and infrared images for testing. Is this operation correct?
- In addition, if I use pix2pix method, whether the following color training and test commands are correct:
1. python train.py --dataroot ./single/colorization_IR --name IR_colorization --model colorization --load_size 2560 --preprocess crop --crop_size 640 --batch_size 2 --netG resnet_9blocks --gpu_ids 0,1,2,3,4,5,6,7
2. python test.py --dataroot ./single/colorization_IR/ --name IR_colorization --model colorization --dataset_mode colorization --preprocess none --netG resnet_9blocks
Look forward to your reply!
It seems that your problem setting is quite different from colorization in which the goal is to predict ab channels given the L channel. If you have paired datasets, I will just use --model pix2pix
and set --input_nc
and --output_nc
properly.
@junyanz thanks, for your reply.
Now I use the pix2pixHD method to verify, and the effect is very good, but the scene is very single. I have a problem: I want to convert IR to RGB, and the scene is traffic. Because the distribution of different traffic roads is certainly different, and I can't have all the traffic scene data sets for training, so what do I need to do to ensure better results?