search

ykdai / BasicPBC

Official Implementation of "Learning Inclusion Matching for Animation Paint Bucket Colorization"
Other
259 stars 23 forks source link

KeyError: "No object named 'PaintBucketLabelSegDataset' found in 'dataset' registry!" #25

Closed riderexin closed 4 months ago

riderexin commented 4 months ago

I run the following command.

"python basicsr/train.py -opt options/train/basicpbc_pbch_train_option.yml"

And an error occurred.

image

I wonder how to find or register 'PaintBucketLabelSegDataset'.

ykdai commented 4 months ago

Thank you for your feedback. You could run python setup.py develop to solve this problem.

riderexin commented 4 months ago

Thank you for your feedback. You could run python setup.py develop to solve this problem.

Thanks for the reply. I had run the "python setup.py develop" already. After testing these days, I found it might be the problem of the torch version. I tried to use torch 2.3.1 before. And the code just asked me to install basicsr and needed a high version of torch-scatter which needs c++14. When I turn to torch1.12.1+cu11.3, things seem to work.

riderexin commented 4 months ago

Thank you for your feedback. You could run python setup.py develop to solve this problem.

By the way, in the paper, it is mentioned that "During the training stage, our model undergoes 300,000 iterations with a batch size of 2, utilizing NVIDIA GeForce RTX 3090 GPUs." would you mind sharing the details of your computing devices? I wonder the GPU RAM and the number of GPUs.

I'm currently running the code on a single 3080ti 12GB. I think the GPU RAM is not big enough to train with a batch size of more than 1.

ykdai commented 4 months ago

We are using NVIDIA GeForce RTX 3090 GPUs as stated in the paper with 24G memory. image