自制2D卷积代码优化

Question

我在python中为CNN im编程编写了以下代码。它不是非常有效，我在跨度为1的512x512图像上运行5x5内核，没有填充，整个卷积需要5-6秒的时间。是否有人对此有优化的解决方案，有效的替代方案（模块或软件包）或其他有关减少运行时间的建议？下面的代码

    y = np.random.uniform(low = 0, high = 5, size = (5, 5))
    x = np.random.uniform(low = 0.1, high = 3, size = (512, 512))

    res = np.zeros(((x.shape[0]-y.shape[0]+1),(x.shape[0]-y.shape[0]+1)), dtype = "float16")



    k = y.shape[0]
    m = x.shape[0]
    n = x.shape[1]
    for i in range(m-(k-1)):
        for j in range((n-(k-1))):
            sum = 0
            for ki in range(k):
                for kj in range(k):
                    data = x[i+ki][j+kj]
                    kval = y[ki][kj]
                    sum = sum + data*kval

            res[i][j] = sum
    print(res)