ValueError: too many values to unpack (expected 3)

[wwzjer]

Dear author,

thanks for sharing the source code of your brilliant paper! When I was trying to run through the pre-training code with 1.6K CT volumes, I met a bug in the line:
img, labels, crops = batch
with ValueError: too many values to unpack (expected 3)
After that, I dig into the "batch" and found it is comprised of dictionaries with keys "image", "foreground_start_coord", "foreground_end_coord". Are these three corresponding to the "img, labels, crops" in your code? Please point out if I made a mistake.

Thank you so much in advance!

[Luffy03]

Hi, many thanks for your attention to our work!
Seems that you did not use the 'voco_augmentation' transform.

VoCo/utils/data_utils.py

Line 87 in 94ed426

VoCoAugmentation(args, aug=True)

It will change the original dict to 'img, labels, crops', you can refer to https://github.com/Luffy03/VoCo/blob/main/utils/data_utils.py.

[wwzjer]

Thanks for your quick reply!
I also found that VoCoAugmentation's function is to generate the 'img, labels, crops'. But I think I did not turn it off when I was running. I have only changed "get_loader(args)" as "get_loader_1k(args)" in the voco_train.py and changed the directory of dataset in the data_utils.py. Could you help me diagnose the issue please? Thanks!

[Luffy03]

It is weird. Would you please provide your 'data_utils.py' file?

[Luffy03]

And do you have problems with the 10k pre-training?

[wwzjer]

Thanks for the suggestions. I have tried to use get_loader instead of get_loader_1k and the issue did not appear. I think it is due to the difference of train_transforms in two functions. But anyway, I think it is fixed.
BTW, I was wondering how many GPU memory do you totally use for pertaining. Thanks!

[Luffy03]

Great to hear that!
Currently, we use one H800 (80G) GPU for training 10k CT.