迁移pytorch工程至matlab
好久没写博客了,险些以为自己找不到密码了。
最近抽空参与了个小项目,很惭愧,只做了三件小事
1. 基于PyTorch训练了一系列单图像超分辨神经网络
基于PyTorch训练了一系列单图像超分辨神经网络,超分辨系数从2-10。
该部分的实现参考了pytorch官方repo中的SR例程,训练程序包含于`./train`文件夹。该项目
基于高效子像素卷积层[1]进行空间分辨率提升操作,训练速度极快。
[1] ["Shi W, Caballero J, Huszar F, et al. Real-Time Single Image and
Video Super-Resolution Using an Efficient Sub-Pixel Convolutional
Neural Network[J]. 2016:1874-1883.](https://arxiv.org/abs/1609.05158)
2. 把训练好的模型权值转存为MATLAB文件。
简单粗暴,异常直接,只要把对应卷积层的权值全部提取出来就可以了。
提取的时候注意一点,要把pytorch中的Variable格式转换为Tensor,再转换为CPU模式,最终转换为numpy数组。
这一系列过程合并起来就是:
Var.data.cpu().numpy()
具体实现如下:
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)
3. 把网络的test过程移植到了MATLAB平台,并撰写了测试代码。
把卷积层和pixelshuffle层用matlab重写了一下。
复现pixelshuffle层的时候遇到了一些麻烦,又回头看了下pytorch里的测试代码
`https://github.com/pytorch/pytorch/blob/master/test/test_nn.py `
# 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])
理了理思路,改写成MATLAB代码:
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
4. 完整工程github链接。
https://github.com/JiJingYu/super-resolution-by-subpixel-convolution
模型权值已保存为matlab权值,直接在matlab中运行`demo.m`文件即可验证