• 大小: 12KB
    文件类型: .zip
    金币: 1
    下载: 0 次
    发布日期: 2021-05-13
  • 语言: Python
  • 标签: python  

资源简介

主要是用于去除雨点的代码,效果非常好,是用python写的

资源截图

代码片段和文件信息

import torch
import torch.nn.functional as F
from torch.autograd import Variable
import numpy as np
from math import exp

def gaussian(window_size sigma):
    gauss = torch.Tensor([exp(-(x - window_size//2)**2/float(2*sigma**2)) for x in range(window_size)])
    return gauss/gauss.sum()

def create_window(window_size channel):
    _1D_window = gaussian(window_size 1.5).unsqueeze(1)
    _2D_window = _1D_window.mm(_1D_window.t()).float().unsqueeze(0).unsqueeze(0)
    window = Variable(_2D_window.expand(channel 1 window_size window_size).contiguous())
    return window

def _ssim(img1 img2 window window_size channel size_average = True):
    mu1 = F.conv2d(img1 window padding = window_size//2 groups = channel)
    mu2 = F.conv2d(img2 window padding = window_size//2 groups = channel)

    mu1_sq = mu1.pow(2)
    mu2_sq = mu2.pow(2)
    mu1_mu2 = mu1*mu2

    sigma1_sq = F.conv2d(img1*img1 window padding = window_size//2 groups = channel) - mu1_sq
    sigma2_sq = F.conv2d(img2*img2 window padding = window_size//2 groups = channel) - mu2_sq
    sigma12 = F.conv2d(img1*img2 window padding = window_size//2 groups = channel) - mu1_mu2

    C1 = 0.01**2
    C2 = 0.03**2

    ssim_map = ((2*mu1_mu2 + C1)*(2*sigma12 + C2))/((mu1_sq + mu2_sq + C1)*(sigma1_sq + sigma2_sq + C2))

    if size_average:
        return ssim_map.mean()
    else:
        return ssim_map.mean(1).mean(1).mean(1)

class SSIM(torch.nn.Module):
    def __init__(self window_size = 11 size_average = True):
        super(SSIM self).__init__()
        self.window_size = window_size
        self.size_average = size_average
        self.channel = 3
        self.window = create_window(window_size self.channel)

    def forward(self img1 img2):
        (_ channel _ _) = img1.size()

        if channel == self.channel and self.window.data.type() == img1.data.type():
            window = self.window
        else:
            window = create_window(self.window_size channel)
            
            if img1.is_cuda:
                window = window.cuda(img1.get_device())
            window = window.type_as(img1)
            
            self.window = window
            self.channel = channel


        return _ssim(img1 img2 window self.window_size channel self.size_average)

def ssim(img1 img2 window_size = 11 size_average = True):
    (_ channel _ _) = img1.size()
    window = create_window(window_size channel)
    
    if img1.is_cuda:
        window = window.cuda(img1.get_device())
    window = window.type_as(img1)
    
    return _ssim(img1 img2 window window_size channel size_average)

 属性            大小     日期    时间   名称
----------- ---------  ---------- -----  ----
     目录           0  2018-07-27 07:55  RESCAN-master\
     文件        1203  2018-07-27 07:55  RESCAN-master\.gitignore
     文件        1065  2018-07-27 07:55  RESCAN-master\LICENSE
     文件        3686  2018-07-27 07:55  RESCAN-master\README.md
     目录           0  2018-07-27 07:55  RESCAN-master\config\
     文件        2635  2018-07-27 07:55  RESCAN-master\config\cal_ssim.py
     文件          96  2018-07-27 07:55  RESCAN-master\config\clean.sh
     文件        4336  2018-07-27 07:55  RESCAN-master\config\dataset.py
     文件        7086  2018-07-27 07:55  RESCAN-master\config\main.py
     文件        6194  2018-07-27 07:55  RESCAN-master\config\model.py
     文件         742  2018-07-27 07:55  RESCAN-master\config\settings.py
     文件        2605  2018-07-27 07:55  RESCAN-master\config\show.py
     文件         233  2018-07-27 07:55  RESCAN-master\config\tensorboard.sh
     目录           0  2018-07-27 07:55  RESCAN-master\docs\
     文件           1  2018-07-27 07:55  RESCAN-master\docs\index.html
     文件         180  2018-07-27 07:55  RESCAN-master\explore.sh

评论

共有 条评论