Closed mrlogick closed 4 years ago
Have you tried redownloading the vgg weights?
On Mon, Feb 3, 2020, 10:02 PM Fabio Vianello notifications@github.com wrote:
I have no issues with stylize.py (with pretrained models). My problem is with train.py. I think I followed the instructions: downloaded vgg16-00b39a1b.pth, downloaded train2014, installed pytorch, opencv, etc (installed with conda). I have Debian 11 stable, Python 3.7.4, pytorch 1.4.0, torchvision 0.5.0, cudatoolkit 10.1.243.
$ python train.py
Traceback (most recent call last): File "train.py", line 168, in
train() File "train.py", line 50, in train VGG = vgg.VGG16().to(device) File "/home/mrfabiolo/style_transfer/fast-neural-style-pytorch/vgg.py", line 33, in init vgg16_features.load_state_dict(torch.load(vgg_path), strict=False) File "/home/mrfabiolo/miniconda3/lib/python3.7/site-packages/torch/serialization.py", line 529, in load return _legacy_load(opened_file, map_location, pickle_module, pickle_load_args) File "/home/mrfabiolo/miniconda3/lib/python3.7/site-packages/torch/serialization.py", line 692, in _legacy_load magic_number = pickle_module.load(f, pickle_load_args) _pickle.UnpicklingError: invalid load key, '<'. Same problem with device = "cpu" or device = "cuda". I got a GeForce GTX 750 card.
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/rrmina/fast-neural-style-pytorch/issues/10?email_source=notifications&email_token=AB3NRXJOWZTPFVKQQICZ3NDRBAIX3A5CNFSM4KPEZMEKYY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4IKSXDVA, or unsubscribe https://github.com/notifications/unsubscribe-auth/AB3NRXJACJFRHWE3VK47GW3RBAIX3ANCNFSM4KPEZMEA .
Oh, i'm stupid. I simply clicked "save as" on the .pth link. But if I open it on the browser I get:
<Error><Code>AccessDenied</Code><Message>Access Denied</Message><RequestId>E8CF943BCC893DFE</RequestId><HostId>8L/NToTygqd2LiDk2yhuGpgWphBTLhE0BCEvpc+AgBbroAQSvCVSO9Gm9hcbNh3pXftyg1HmJ78=</HostId></Error>
Do I need to sign up to amazonaws.com or something? Is it free?
Seems like Justin Johnson has already stopped hosting the vgg weights. I've uploaded it in my Google Drive. Here's the link: https://drive.google.com/file/d/1a0sFcNEvmIy21PE0yp7tzJuU0vhhV0Ln/view?usp=sharing
Let me know if it works or not.
Yes! It works, thanks man
I have no issues with stylize.py (with pretrained models). My problem is with train.py. I think I followed the instructions: downloaded vgg16-00b39a1b.pth, downloaded train2014, installed pytorch, opencv, etc (installed with conda). I have Debian 11 stable, Python 3.7.4, pytorch 1.4.0, torchvision 0.5.0, cudatoolkit 10.1.243.
Same problem with
device = "cpu"
ordevice = "cuda"
. I got a GeForce GTX 750 card.