验证torch.nn.Conv2d

import os
import cv2
import torch
import numpy as np
import random
import cv2 as cv
from matplotlib import pyplot as plt

def f_VerifyConv2D():
    """
        验证torch.nn.Conv2d， 并将输入数据及权重保存到txt文件中
    """
    x = torch.randn(1, 1, 10, 10)
    x = x.round().float()

    print('==========================================================================')
    print('================================== 输入数据 ')
    print(x)
    kernel_size = 1
    conv_zeros = torch.nn.Conv2d(in_channels=1, out_channels=3, kernel_size=kernel_size, stride=1, padding=0, bias=True)
    conv_zeros.weight = torch.nn.Parameter(torch.ones(1, 1, kernel_size, kernel_size))
    conv_zeros.bias = torch.nn.Parameter(torch.ones(1))
    y = conv_zeros(x)
    print('==========================================================================')
    print('================================== 卷积权重数据 ')
    print(conv_zeros.state_dict())
    print('==========================================================================')
    print('================================== 卷积输出数据 ')
    print(y)

    print('==========================================================================')
    print('================================== 保存参数txt ')
    # 保存输入数据到1.txt
    ndarray = x.numpy()
    ndarray = ndarray.reshape(-1).astype(np.int16)
    ndarray[np.where(ndarray < 0)] = np.add(ndarray[np.where(ndarray < 0)], 256)  # 将负数以补码的形式进行存储
    np.savetxt("1.txt", ndarray, fmt='%x', delimiter='\n')

    # 保存输出结果到2.txt
    ndarray_o = y.detach().numpy()
    ndarray_o = ndarray_o.reshape(-1).astype(np.int16)
    ndarray_o[np.where(ndarray_o < 0)] = np.add(ndarray_o[np.where(ndarray_o < 0)], 256)  # 将负数以补码的形式进行存储
    np.savetxt("2.txt", ndarray_o, fmt='%x', delimiter='\n')

f_VerifyConv2D()