仿射变换实现letterbox形式的resize

最新推荐文章于 2024-05-31 19:58:58 发布

是大糊涂不聪明

最新推荐文章于 2024-05-31 19:58:58 发布

阅读量437

点赞数

分类专栏：目标检测文章标签：目标检测

本文链接：https://blog.csdn.net/weixin_47289438/article/details/127682948

版权

目标检测专栏收录该内容

6 篇文章 0 订阅

订阅专栏

文章目录

1. yolov5中letterbox实现
2. 仿射变换

为啥会用仿射变换实现，因为高级…

1. yolov5中letterbox实现

保持比例缩放，防止变形。长边缩放
短边填充

import cv2 as cv
import numpy as np
from typing import Optional, Tuple
import matplotlib.pyplot as plt


def letterbox(img, new_shape=(640, 640), color=(114, 114, 114), auto=False, scaleFill=False, scaleup=True, stride=32):
    # Resize and pad image while meeting stride-multiple constraints
    shape = img.shape[:2]  # current shape [height, width]
    if isinstance(new_shape, int):
        new_shape = (new_shape, new_shape)

    # Scale ratio (new / old)
    r = min(new_shape[0] / shape[0], new_shape[1] / shape[1])
    if not scaleup:  # only scale down, do not scale up (for better test mAP)
        r = min(r, 1.0)

    # Compute padding
    ratio = r, r  # width, height ratios
    new_unpad = int(round(shape[1] * r)), int(round(shape[0] * r))
    dw, dh = new_shape[1] - new_unpad[0], new_shape[0] - new_unpad[1]  # wh padding
    if auto:  # minimum rectangle
        dw, dh = np.mod(dw, stride), np.mod(dh, stride)  # wh padding
    elif scaleFill:  # stretch
        dw, dh = 0.0, 0.0
        new_unpad = (new_shape[1], new_shape[0])
        ratio = new_shape[1] / shape[1], new_shape[0] / shape[0]  # width, height ratios

    dw /= 2  # divide padding into 2 sides
    dh /= 2
    dw = int(dw)
    dh = int(dh)
    if shape[::-1] != new_unpad:  # resize
        img = cv.resize(img, new_unpad, interpolation=cv.INTER_LINEAR)
    top, bottom = int(round(dh - 0.1)), int(round(dh + 0.1))
    left, right = int(round(dw - 0.1)), int(round(dw + 0.1))
    img = cv.copyMakeBorder(img, top, bottom, left, right, cv.BORDER_CONSTANT, value=color)  # add border
    return img, ratio[0], (dw, dh)  # letterbox默认等比缩放，取一个值就可

img = cv.imread('./test_img_xml/1.jpg')
dst = letterbox(img)[0]

plt.figure(1)
plt.imshow(img[:, :, ::-1], cmap='gray')
plt.figure(2)
plt.imshow(dst[:, :, ::-1], cmap='gray')
plt.show()

在这里插入图片描述

2. 仿射变换

先将图片平移到左上角原点，得到Center矩阵
再将图片resize到目标尺寸，得到Resize矩阵
再将图片中心由左上角原点平移到缩放后图像的中心点，得到Translate矩阵
最后的仿射变换矩阵 M = T @ R @ C
有需要的话，可以再填充边缘

img = cv.imread('./test_img_xml/1.jpg')
h0, w0 = img.shape[:2]
print(w0, h0)

C = np.eye(3)
C[0, 2] = -w0/2
C[1, 2] = -h0/2
dst_shape = get_minimum_dst_shape((w0, h0), (640, 640))
w1, h1 = dst_shape
print(dst_shape)
R = get_resize_matrix((w0, h0), (w1, h1))
print(R)
T = np.eye(3)
T[0, 2] = w1/2
T[1, 2] = h1/2

M = T @ R @ C
dst = cv.warpAffine(img, M[:2], (w1, h1))


padh = int((640-h1)/2)
padw = int((640 - w1)/2)
dst2 = cv.copyMakeBorder(dst, padh, padh, padw, padw, cv.BORDER_CONSTANT, value=(114, 114, 114))
plt.figure(1)
plt.imshow(img[:, :, ::-1], cmap='gray')
plt.figure(2)
plt.imshow(dst[:, :, ::-1], cmap='gray')
plt.figure(3)
plt.imshow(dst2[:, :, ::-1], cmap='gray')
plt.show()