ktaebum/attentioneddeeppaint

Automatic Colorization of Anime Sketch Image

attentioneddeeppaint's Issues

what a cool stuff！

let me try it

what if, replace your ref image by a color sketch from the artist, can it help drawing?

I mean, if this script can be linked to stable diffusion,
and if im the artist, i draw the line art and I draw the color ref( set up the color theme , like color sketch plan, but not fully render)
then use your script , what will happen?

will it combine the line art and the color theme I set up, and make a final image using stable diffusion, this would be a cool function for
professional artist, i mean, for real.

implementation

• UNet with skip connection and resblock
• Apply catoonGAN’s idea

Seems like the dataset is damaged

When I tried to unzip the kaggle dataset (https://www.kaggle.com/ktaebum/anime-sketch-colorization-pair), I've discovered that the dataset has been damaged.
It doesn't even unzip "val" folder.

I've tried downloading it from different internet, and even tried to download it from the old dataset (https://www.kaggle.com/ktaebum/animesketchcolorpair)

My environment is Windows 10 Home version 1903 for OS
and Bandizip for unzipper.

May you share some pre-trained models?

Since not all people have a decent GPU to train the network (I only have a laptop), I think it's a good idea to share a pre-trained network.
Preferably the one used to make the images on the README.md

deepunet.py raises an error about local variable 'i' reference before assignment

File "/home/alexandrugh/Documents/work/sketchColorization/AttentionedDeepPaint/trainer/deepunet.py", line 136, in train
return i
UnboundLocalError: local variable 'i' referenced before assignment

Any idea or workaround for this?
Thanks a lot and congrats for the great work!
It would be also great to have a series of pretrained models that we can finetune or test on single images.

Can anyone share some of the trained models?

This program looks great,maybe better than paintschainer or s2p
I trying execute the program on the google colab,but get a oom error,
can anyone share some of the trained models?
Thankss a lot.

Why are my results rough？

Your effect is really amazing.
I changed the size of the image to 256.
Is this an intermediate outcome from your training？

7703 Segmentation fault.

Could this be a lack of permission in the Dir? The dataset was corrupted but I was able to unzip most of it.

train.sh: line 3: 7703 Segmentation fault (core dumped) python train.py --verbose --save-every 5 --model deepunet --sample 4 --no-mse --num-epochs 15 --print-every 300 --train

Error in training (num_samples = 0

Traceback (most recent call last):
File "train.py", line 77, in
main(parser.parse_args())
File "train.py", line 43, in main
batch_size=args.batch_size,
File "G:\PYTHON\python3.5.4\lib\site-packages\torch\utils\data\dataloader.py", line 176, in init
sampler = RandomSampler(dataset)
File "G:\PYTHON\python3.5.4\lib\site-packages\torch\utils\data\sampler.py", line 66, in init
"value, but got num_samples={}".format(self.num_samples))
ValueError: num_samples should be a positive integer value, but got num_samples=0

torch version ==1.1.0
OS: windows10

should i specify some value in train.py?

paper

where can i find the corresponding paper?

Cuda Causes Trainer to Hang

Weird question, I have 2 cuda cards and the code only uses one. It runs great on my laptop because its on the CPU. Can you tell me how to have it identify 2nd card or force CPU?

RuntimeError: CUDA out of memory. Tried to allocate 256.00 MiB (GPU 0; 8.00 GiB total capacity; 6.15 GiB already allocated; 58.05 MiB free; 83.38 MiB cached)

I removed cuda from the workstation and works great on cpu

inputs type is sketch image and style image?

Hello, thank you for revealing the source code.
But the checkpoint doesn't seem to be public. If there is a pretrained checkpoint, please link it.

Q1.
Do you have to have a sketch image and a style image unconditionally?

기억에 의존한 purpose

1. 훌륭한 채색
2. GAN에서의 효율적인 evaluation metric