D-net论文复现部分内容 --关于人工特征融入深度网络的思考。

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102

#xioabo.py

#!/usr/bin/env python 
# -*- coding:utf-8 -*-
import numpy as np
import pywt
import cv2
import matplotlib.pyplot as plt
import torch


def xiaob(img):
    img=img.reshape(img.shape[1], img.shape[2],img.shape[3])
    img = img.permute(1, 2, 0)
    img=img.cpu().numpy()  # 把数据集中的tensor转numpy
    # print(type(img2))
    # img3 = cv2.resize(img2, (448, 448))
    b, g, r = cv2.split(img)
    x = (b, g, r)
    count = 1
    for img in x:
        # plt.figure('二维小波一级变换')
        coeffs = pywt.dwt2(img, 'haar')
        cA, (cH, cV, cD) = coeffs

        # 将各个子图进行拼接，最后得到一张图
        '''AH = np.concatenate([cA, cH], axis=1)
        VD = np.concatenate([cV, cD], axis=1)
        img = np.concatenate([AH, VD], axis=0)'''
        # 显示为灰度图
        ''' plt.imshow(img, 'gray')
        plt.title('result')
        plt.show()'''
        cA = torch.from_numpy(cA)
        cH = torch.from_numpy(cH)
        cV = torch.from_numpy(cV)
        cD = torch.from_numpy(cD)
        cA = cA.resize(1, cA.shape[0], cA.shape[1])
        cH = cH.resize(1, cH.shape[0], cH.shape[1])
        cV = cV.resize(1, cV.shape[0], cV.shape[1])
        cD = cD.resize(1, cD.shape[0], cD.shape[1])

        if count == 1:
            # y1 = (cV, cA, cH, cD)
            # cv2.imshow("y1", cD)
            outputs1 = torch.cat((cV, cA, cH, cD), dim=0)
            k=cA
            # print(outputs1.shape)
        if count == 2:
            # y2 = (cV, cA, cH, cD)
            outputs2 = torch.cat((cV, cA, cH, cD), dim=0)
            k = torch.cat((k, cA), dim=0)
        if count == 3:
            # y3 = (cV, cA, cH, cD)
            outputs3 = torch.cat((cV, cA, cH, cD), dim=0)
            k = torch.cat((k, cA), dim=0)
        count += 1
    out = (outputs1, outputs2, outputs3)
    outputs = torch.cat(out, dim=0)
    outputs=outputs.resize(1,outputs.shape[0], outputs.shape[1], outputs.shape[2])
    k=k.resize(1,k.shape[0], k.shape[1], k.shape[2])
    return (k.cuda(),outputs.cuda())  # 12层特征

# 将多通道图像变为单通道图像
# img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY).astype(np.float32)
# img = cv2.imread("C:/Users/brighten/Desktop/1.jpg")
#img = torch.rand(size=(3,224, 224))
#img=img.permute(1, 2, 0)
# print(img.shape)
# print(img)
'''k,y = xiaob(img)
print(k.shape)
print(y.shape)

k,y = xiaob(k)
print(k.shape)
print(y.shape)'''


# cv2.waitKey(0)
'''
(180, 130, 3)
torch.Size([4, 90, 65])
torch.Size([12, 90, 65])

输入rgb 图像，（w,h,3）
输出 tensor张量  (12,w/2,h/2)
这两个表示图层的方式不一样


改后。
输入张量  

意思是把输入的张量按照numpy 的方式放置维度信息
输出张量  


大小同上
'''