Open stphtan94117 opened 1 year ago
Thank you for your interest on our project.
The script 'colorize.real.sh' is set to use a batch size of 1, so changing the batch size will not solve the problem. The amount of GPU memory needed depends on the aspect ratio of the input image because the script uses a resize trick to set the shorter length to 256. When colorizing a square image, it requires about 6GB of memory. I suspect you tried using an image with a larger aspect ratio.There are two options you could try. The first is to crop your image to reduce the aspect ratio. The other option is to reduce the '--size_target' option in the 'colorize.real.sh' script, which is set to 256 by default. Keep in mind that this may degrade the quality of the output.
I have reduced the photo to 160x120pixel, and then changed --size_target to 16, but cuda out of memory still appears.
This is really wierd. A possible scenario I can imagine is that there is an other process which is using a huge parts of the GPU memory. Did you check the remaining usage GPU memory before you run the script?
i have checked the remaining usage GPU memory before I run the script.
only 'colorize.real.sh' is no working.
colorize.multi_c.sh、colorize.multi_z.sh、infer.bigcolor.e011.sh,they are working for my computer. it is so wierd.
I'm sorry it is really hard to know what the problem is. The next steps I can recommend to you are as follow
Or there is a way to conduct inference through web. https://replicate.com/cjwbw/bigcolor
I'm sorry it is really hard to know what the problem is. The next steps I can recommend to you are as follow
- Check the version of Pytorch. v1.10.0
- Use Ubuntu 20.04.2 LTS operation system
Or there is a way to conduct inference through web. https://replicate.com/cjwbw/bigcolor
my Pytorch is v1.10, but operation system is windows 10. so windows 10 can not inference it? I don't think it has anything to do with the operating system(win10), because several other .sh files can execute.
I have used that website but I am not satisfied with the results. I use deoldify, no matter the speed or coloring is good,but it cannot train.
can you write a code to put on colab?
Maybe try running the device in cup mode (see if it works) --device=cpu
I was --topk default=5, but I couldn't succeed, and a blank paper came out. Changing to topk default=1 is ok, but the color is weird.
(^~^;)ゞ May I ask what aspect is the setting of topk related to?
name = path.split('/')[-1].split('.')[0] is always wrong. Changed to name =os.path.basename(path) is finally OK.
(Ɔ˘⌣˘)(˘⌣˘)˘⌣˘ C)
hi sir my GPU is nvidia RTX3080, its cuda memory is 10 GB i run "sh ./scripts/colorize.real.sh" it will show cuda out of memory..... Where can I modify the batch or size?
thanks