Pytorch code training problem about lightcnn HOT 4 CLOSED

oop! I can't use the code to train the Celeb datasets either. But my error not like this. The error is:
Traceback (most recent call last):
File "train.py", line 288, in
main()
File "train.py", line 95, in main
weight_decay=args.weight_decay)
File "/home/hezhenhao/.conda/envs/my_root/lib/python3.6/site-packages/torch/optim/sgd.py", line 56, in init
super(SGD, self).init(params, defaults)
File "/home/hezhenhao/.conda/envs/my_root/lib/python3.6/site-packages/torch/optim/optimizer.py", line 61, in init
raise ValueError("can't optimize a non-leaf Variable")
ValueError: can't optimize a non-leaf Variable

Hope someone can help me fix the issue.

from lightcnn.

@NIIAS3050 Refer to issue #12 and #4

@scoyer Refer to #1 or update your pytorch to the latest version

from lightcnn.

Hi again.
In your paper you wrote, that xavier and gaussian inintailization was used .
But in your Python code there is no weight initialization. In your answer in #12 you reccomend to use pretrained model as initialization. Can i use your python code to learn LightCNN from scratch?
Or it can be used only for finetuning(without modifications of course)?

from lightcnn.

You can learn LightCNN from scratch. I try it after I change "shuffle = True" during loading the dataset and it works well. And I think you better modify some hyper parameters if the training dataset is not MS-cele.

from lightcnn.