szagoruyko/diracnets

Training Very Deep Neural Networks Without Skip-Connections

diracnets's Issues

What's means to Dirac parameterization?

I want to input my own data.

Hi, @soumith @szagoruyko

I want to input my own data in your diracNet.
Not a Cifar-10/Imagenet data but my own images data folder.

Like this code: http://pytorch.org/tutorials/beginner/transfer_learning_tutorial.html

Is that possible?

Thanks in advance ~

Some codes may be different from what you said in you paper

Hi, Thank you for your code. But I am confused about the "i" implemented in your code. I think it may be different from what you said in your paper. In your paper, you initialize "I" as :

In some filters, there may be more than one weight initialized as 1. But in your code:
dirac_(I)
you used dirac_ function to initialize it. Only one element in a filter could be set 1. Maybe I misunderstand it. Could you please help me with it?

Can not open the pretrained file

there are four files that you supply,but the last two files I can not download it.It display error message

no torchnet

how to download and install the torchnet module ? in sentence: import torchnet as tnt

requirements.txt?

The documentation mentions installing from a requirements.txt with pip install -r requirements.txt but there's no requirements.txt present in this repository.

Can not run train.py script.

I run python train --help, and get one error:

$ python train.py --help
File "train.py", line 166
    z = {**vars(opt), **t}
          ^
SyntaxError: invalid syntax

How could I fix this error?

invalid combination of arguments when run the tran.py

raceback (most recent call last):
File "train.py", line 230, in
main()
File "train.py", line 118, in main
f, params, stats = define_diracnet(opt.depth, opt.width, opt.dataset)
File "/home/amax/cqs/diracnets/diracnet.py", line 126, in define_diracnet
'conv': cast(kaiming_normal(torch.Tensor(widths[0], 3, 3, 3))),
TypeError: new() received an invalid combination of arguments - got (Tensor, int, int, int), but expected one of:

• (torch.device device)
• (tuple of ints size, torch.device device)
• (torch.Storage storage)
• (Tensor other)
• (object data, torch.device device)

AttributeError: 'module' object has no attribute 'normalize'

Hi, when I train the network, I got an error as followed:

recent call last):
File "train.py", line 230, in
main()
File "train.py", line 226, in main
engine.train(h, train_loader, opt.epochs, optimizer)
File "/home/jlin/anaconda2/envs/pytorch/lib/python2.7/site-packages/torchnet/engine/engine.py", line 63, in train
state['optimizer'].step(closure)
File "/home/jlin/anaconda2/envs/pytorch/lib/python2.7/site-packages/torch/optim/sgd.py", line 72, in step
loss = closure()
File "/home/jlin/anaconda2/envs/pytorch/lib/python2.7/site-packages/torchnet/engine/engine.py", line 52, in closure
loss, output = state'network'
File "train.py", line 162, in h
y = data_parallel(f, inputs, params, stats, sample[2], list(np.arange(opt.ngpu)))
File "/home/jlin/diracnets-master/diracnet.py", line 44, in data_parallel
return f(input, params, stats, mode)
File "/home/jlin/diracnets-master/diracnet.py", line 116, in f
o = group(o, params, stats, 'group0', mode, n * 2)
File "/home/jlin/diracnets-master/diracnet.py", line 94, in group
o = block(o, params, stats, '%s.block%d' % (base, i), mode, i)
File "/home/jlin/diracnets-master/diracnet.py", line 86, in block
w = beta * F.normalize(w.view(w.size(0), -1)).view_as(w) + alpha * delta
AttributeError: 'module' object has no attribute 'normalize'

And I view the file torch.nn.functional, where I could not find the functional named normalize. Did anyone meet the same problem ?

How Can I run this code in CPU only mode?

HI, @soumith @szagoruyko.

How Can I run this code in CPU only mode?

Is that Possible ?

Thanks.

cublas runtime error

On ImageNet set, it occured as following:
File "train.py", line 245, in
main()
File "train.py", line 241, in main
engine.train(h, train_loader, opt.epochs, optimizer)
File "build/bdist.linux-x86_64/egg/torchnet/engine/engine.py", line 39, in train
File "/usr/local/lib/python2.7/dist-packages/torch/optim/sgd.py", line 72, in step
loss = closure()
File "build/bdist.linux-x86_64/egg/torchnet/engine/engine.py", line 28, in closure
File "train.py", line 177, in h
y = data_parallel(f, inputs, params, stats, sample[2], np.arange(opt.ngpu))
File "/home/yq/work/face_class/diracnets/diracnet.py", line 51, in data_parallel
return f(input, params, stats, mode)
File "/home/yq/work/face_class/diracnets/diracnet.py", line 182, in f
o = F.linear(o.view(o.size(0), -1), params['fc.weight'], params['fc.bias'])
File "/usr/local/lib/python2.7/dist-packages/torch/nn/functional.py", line 449, in linear
return state(input, weight) if bias is None else state(input, weight, bias)
File "/usr/local/lib/python2.7/dist-packages/torch/nn/functions/linear.py", line 10, in forward
output.addmm(0, 1, input, weight.t())
RuntimeError: cublas runtime error : library not initialized at /b/wheel/pytorch-src/torch/lib/THC/THCGeneral.c:394

Why DiracNet(18) or DiracNet(34) doesn't have DiracConv ?

TypeError: float() argument must be a string or a number

Traceback (most recent call last):
File "train.py", line 231, in
main()
File "train.py", line 227, in main
engine.train(h, train_loader, opt.epochs, optimizer)
File "/home/jlin/anaconda2/envs/pytorch/lib/python2.7/site-packages/torchnet/engine/engine.py", line 63, in train
state['optimizer'].step(closure)
File "/home/jlin/anaconda2/envs/pytorch/lib/python2.7/site-packages/torch/optim/sgd.py", line 72, in step
loss = closure()
File "/home/jlin/anaconda2/envs/pytorch/lib/python2.7/site-packages/torchnet/engine/engine.py", line 56, in closure
self.hook('on_forward', state)
File "/home/jlin/anaconda2/envs/pytorch/lib/python2.7/site-packages/torchnet/engine/engine.py", line 31, in hook
self.hooksname
File "train.py", line 180, in on_forward
meter_loss.add(float(state['loss']))
TypeError: float() argument must be a string or a number

When I train the network, I met a error like that. Did anyone have meet the same problem, and how to solve it?

no torch.nn.functional.normalize

AttributeError: 'module' object has no attribute 'normalize'

File "/home/yq/work/face_class/diracnets/diracnet.py", line 97, in block
w = beta * F.normalize(w.view(w.size(0), -1)).view_as(w) + alpha * delta