Questions about transfer time

[FantasyJXF]

I use the cmd python WCT.py to generate the image, and find each image costs more than 40 seconds, I use my MAC(Retina, 15-inch, Mid 2015) to run the model.

Your 12GB TITAN X costs less than 1 second? I read your paper, it said so.

Elapsed time is: 44.224898
Transferring in3.jpg
Elapsed time is: 43.248152
Transferring in1.jpg
Elapsed time is: 41.960227
Transferring in4.jpg
Elapsed time is: 51.534320
Processed 4 images. Averaged time is 45.241899

[liwei92]

Using the GPU.

[FantasyJXF]

Another question:
Why I get different style transfer results?
I download the models you provide, and run the python WCT.py with pytorch v0.4.0 installed.

I think you may be using v0.3.0

The left was mine, the right yours.

Does the version matters?

Your results are much more better.

[FantasyJXF]

I try to use PyTorch 0.3.0, there is no warning, the result is the same as my PyTorch 0.4.0.

Hope your reply

[sunshineatnoon]

The results are produced by using 4 auto-encoders, I guess you use 5 auto-encoders.

[FantasyJXF]

I check the code, maybe that's the problem, I'm going to try.

Thanks

class WCT(nn.Module):
    def __init__(self,args):
        super(WCT, self).__init__()
        # load pre-trained network
        vgg1 = load_lua(args.vgg1)
        decoder1_torch = load_lua(args.decoder1)
        vgg2 = load_lua(args.vgg2)
        decoder2_torch = load_lua(args.decoder2)
        vgg3 = load_lua(args.vgg3)
        decoder3_torch = load_lua(args.decoder3)
        vgg4 = load_lua(args.vgg4)
        decoder4_torch = load_lua(args.decoder4)
        vgg5 = load_lua(args.vgg5)
        decoder5_torch = load_lua(args.decoder5)


        self.e1 = encoder1(vgg1)
        self.d1 = decoder1(decoder1_torch)
        self.e2 = encoder2(vgg2)
        self.d2 = decoder2(decoder2_torch)
        self.e3 = encoder3(vgg3)
        self.d3 = decoder3(decoder3_torch)
        self.e4 = encoder4(vgg4)
        self.d4 = decoder4(decoder4_torch)
        self.e5 = encoder5(vgg5)
        self.d5 = decoder5(decoder5_torch)