Var.data.cpu().numpy()

 1 from __future__ import print_function
 2 
 3 import torch
 4 import numpy as np
 5 import scipy.io as sio
 6 
 7 for i in [2, 3, 4, 5, 6, 7, 8, 9, 10]:
 8 
 9     model_name = 'model_upscale_{}_epoch_101.pth'.format(i)
10     model = torch.load(model_name)
11     print(model._modules)
12 
13     weight = dict()
14     weight['conv1_w'] = model._modules['conv1']._parameters['weight'].data.cpu().numpy()
15     weight['conv2_w'] = model._modules['conv2']._parameters['weight'].data.cpu().numpy()
16     weight['conv3_w'] = model._modules['conv3']._parameters['weight'].data.cpu().numpy()
17     weight['conv4_w'] = model._modules['conv4']._parameters['weight'].data.cpu().numpy()
18 
19     weight['conv1_b'] = model._modules['conv1']._parameters['bias'].data.cpu().numpy()
20     weight['conv2_b'] = model._modules['conv2']._parameters['bias'].data.cpu().numpy()
21     weight['conv3_b'] = model._modules['conv3']._parameters['bias'].data.cpu().numpy()
22     weight['conv4_b'] = model._modules['conv4']._parameters['bias'].data.cpu().numpy()
23 
24     sio.savemat('model_upscale_{}.mat'.format(i), mdict=weight)

# https://github.com/pytorch/pytorch/blob/master/test/test_nn.py
def _verify_pixel_shuffle(self, input, output, upscale_factor):
    for c in range(output.size(1)):
        for h in range(output.size(2)):
            for w in range(output.size(3)):
                height_idx = h // upscale_factor
                weight_idx = w // upscale_factor
                channel_idx = (upscale_factor * (h % upscale_factor)) + (w % upscale_factor) + \
                              (c * upscale_factor ** 2)
    self.assertEqual(output[:, c, h, w], input[:, channel_idx, height_idx, weight_idx])

 1 function [ outputs ] = PixelShuffle( inputs, upscale_factor )
 2 %    PixelShuffle :
 3 %
 4 %   input : N, upscale_factor ** 2, H, W
 5 %   output : N, 1, H*upscale_factor, W*upscale_factor
 6 
 7 [N, ~, H, W] = size(inputs);
 8 H_out = H*upscale_factor;
 9 W_out = W*upscale_factor;
10 outputs = zeros([N, 1, H_out, W_out]);
11 for i = 1:N
12     for h = 1: H_out
13         for w = 1:W_out
14             height_idx = floor(h / upscale_factor+0.5);
15             weight_idx = floor(w / upscale_factor+0.5);
16             channel_idx = (upscale_factor * mod(h-1, upscale_factor)) + mod(w-1, upscale_factor)+1;
17             outputs(i, 1, h, w) = inputs(i, channel_idx, height_idx, weight_idx);
18         end
19     end
20 end
21 end