Closed ramdhan1989 closed 2 years ago
Hey @ramdhan1989 thanks for your comments abou the library, really appreciate it.
About Dali, I'm almost sure it's not supported on windows. I would suggest you to check their repo just to confirm, but maybe it's in their roadmap for the future.
About the second part, I'll look into it in the next few days and back to you. I'm writing just to let you know that I saw your issue.
Thank you for your help. I tried using cifar10 dataset but ended up with the same error message.
Can you share the command you ran for cifar10? I think you are just missing the val_dir parameter.
this is my command
python main_pretrain.py --dataset cifar10 --backbone resnet18 --data_dir ./datasets --num_workers 4 --precision 16 --optimizer sgd --lars --grad_clip_lars --eta_lars 0.02 --exclude_bias_n_norm --scheduler warmup_cosine --lr 0.3 --weight_decay 1e-4 --batch_size 256 --brightness 0.4 --contrast 0.4 --saturation 0.2 --hue 0.1 --gaussian_prob 0.0 --solarization_prob 0.0 --name barlow-cifar10 --project self-superivsed --wandb --save_checkpoint --method barlow_twins --proj_hidden_dim 2048 --scale_loss 0.1 --devices 0 --accelerator gpu --max_epochs 1000
Sorry about the delay. I tried to run your command and it worked just fine. There is some stuff missing, e.g., the crop_size (for cifar) which will make images super large for cifar, but apart from that, I got no errors. Did you try to update the repo?
Closing because I didn't manage to reproduce. Feel free to reopen if you have any extra info.
Hi there! this is awesome library and the most complete so far. I would like to train custom dataset. I have both unlabeled and labeled dataset. The labeled dataset is divided into train and validation where the name of folder inside is the class name. I use windows with one gpu. I tried to install dali but it ended up with an error as follow :
ERROR: No matching distribution found for nvidia-dali-cuda110
do you have any idea to solve this?Then I tried not to use dali by using command below:
python main_pretrain.py --backbone resnet18 --data_dir D:/Ramdhan/dynocard/dyno_image_floodfill_after_inv_corr --train_dir dyno_image_floodfill_after_inv_corr --name barlow-custom --project self --method barlow_twins --no_labels --dataset custom --brightness 0.4 --contrast 0.4 --saturation 0.2 --hue 0.1 --optimizer sgd --devices 0 --accelerator gpu --auto_select_gpus --scheduler warmup_cosine --max_epochs 50
Then, it gave an error below. I think the val_loader is None as a result of this line https://github.com/vturrisi/solo-learn/blob/main/main_pretrain.py#:~:text=if%20args.dataset,val_loader%20%3D%20None
please advise thank you