pytorch bilinear interpolation

1、单线性插值

化简得：
重要公式

将yTreated as a function of pixel values;

2、双线性插值

问题：求P点的像素值？

According to the unilinear interpolation formula：

1、得到R1和R2点的像素值：

2、然后通过R1和R2线性插值得到P点的像素值：

所以,A total of cubic unilinear interpolation is used,final pixel value.

另外,Which is due to the difference between adjacent pixels1,所以y2 - y1 = 1 ,和x2-x1 = 1,So the denominator is 1.

最终得到的计算公式为：

3、最近邻法

使用下面公式,Find the nearest pixel value

其中：

存在问题：右偏移

The original formula is offset right,The new formula centers on it.

Because it is offset right in the original formula,So use the center point coincidence to eliminate it.

Below is the optimization formula：

4、Simple implementation of bilinear interpolation

Find by nearest neighborP点,Then you need to find four adjacent pixels.

通过floorThe function finds the lower bound,floor +1 find the upper limit,But to prevent exceeding the pixel coordinate value of the image

import numpy as np
import matplotlib.pyplot as plt
from PIL import Image


def Bilinear(dst,des_w,des_h,src_w,src_h):

    for c in range(3):
        for dst_x in range(des_w):
            for dst_y in range(des_h):
                src_x = (dst_x + 0.5)*src_w/des_w - 0.5
                src_y = (dst_y + 0.5)*src_h/des_h - 0.5

                #Four proximity points
                src_x_1 = int(np.floor(src_x))
                src_y_1 = int(np.floor(src_y))
                src_x_2 = min(src_x_1 + 1,src_w -1)
                src_y_2 = min(src_y_1 + 1,src_h -1)

                R1 = (src_x_2 - src_x) * src[src_y_1,src_x_1,c] + (src_x - src_x_1) * src[src_y_1,src_x_2,c]
                R2 = (src_x_2 - src_x) * src[src_y_2,src_x_1,c] + (src_x - src_x_1) * src[src_y_2,src_x_2,c]
                P = int((src_y_2 - src_y) * R1 + (src_y - src_y_1) * R2)
                dst[dst_y, dst_x, c] = P

    return dst


def show_img(dst):
    dst = dst.astype(np.uint8)
    plt.figure()
    plt.subplot(121)
    plt.imshow(src)
    plt.subplot(122)
    plt.imshow(dst)
    # plt.imsave("./img.png",dst)
    plt.show()


if __name__ == '__main__':

    src = Image.open("./img_1.png")
    src_w = src.width
    src_h = src.height
    src = np.array(src)

    dst = np.ones((960, 1280, 3))
    des_w = dst.shape[1]
    des_h = dst.shape[0]
    # print(des_w,des_h)

    dst = Bilinear(dst,des_w,des_h,src_w,src_h)
    show_img(dst)

5、pytorch中双线性插值

import torch
from torch.nn import functional as F
from PIL import Image
import numpy as np
import matplotlib.pyplot as plt

img = Image.open("./img.png")
img = np.array(img,dtype=float)
print(img.shape)
img = torch.from_numpy(img)
print(img.shape)
img = img.unsqueeze(0).permute(0,3,1,2) #[b,c,h,w]

img = F.interpolate(img,scale_factor=(2,2),mode='bilinear')
# print(img.shape)
img = img.squeeze(0).permute(1,2,0)
print(img.shape)

a = torch.tensor(img, dtype=torch.uint8)
print(a.shape)
plt.figure()
plt.imshow(a)
plt.show()