Open MuadDev opened 3 years ago
Dear Luis,
Thank you for making this code publicly available. I do encounter an error while trying to train on the CEDAR dataset though, perhaps you could help me?
The error looks like this:
The script I use to start training looks like this:
python3.7 -m sigver.featurelearning.train \ --dataset cedar \ --model signet \ --dataset-path /data/signature_matching/data/processed/sigver_datasets/cedar.npz \ --users 10 20 \ --epochs 2 \ --logdir /tmp/signet
The cedar.npz dataset has been preprocessed as has been indicated in the README.
Hope you can help me with this.
The problem seems to be the combination of the parameters "epochs" and "lr_decay_times" (this one defaults to 3).
For this project, I used a step decay for the learning rate (like the original resnet papers), with a default of 3 decays. So if you train with, say 60 epochs, it would decay at 20, 40 and 60. The problem here is that you are asking to train for 2 epochs, and using the default of 3 decays, which won't work: this line: In line https://github.com/luizgh/sigver/blob/master/sigver/featurelearning/train.py#L71, will tell the scheduler to change the learning rate each 0 epochs, which won't work.
So your options are: 1) increasing the number of epochs; 2) Decreasing "lr_decay_times", 3) Changing the learning rate scheduler (nowadays I use cosine annealing for my projects: https://pytorch.org/docs/stable/generated/torch.optim.lr_scheduler.CosineAnnealingLR.html)
Dear Luis,
Thank you for making this code publicly available. I do encounter an error while trying to train on the CEDAR dataset though, perhaps you could help me?
The error looks like this:![image](https://user-images.githubusercontent.com/29200372/110650167-9940e800-81ba-11eb-8314-6de4fb424575.png)
The script I use to start training looks like this:
The cedar.npz dataset has been preprocessed as has been indicated in the README.
Hope you can help me with this.