前言

我不理解为啥opencv-python可视化一堆报错，同一个三通道图像，cv2.imshow()没有问题，cv2.circle()就一直有问题，搞了一晚，心态炸了！！！

cv2需要的图片矩阵（H, W, C）有两种

1. 像素值范围为 [0,1] ------ 数据类型： float32
2. 像素值范围为 [0,255] ------ 数据类型： uint8

源代码

        for k, xys1 in enumerate(output_reproj['keyp_2d']):
            new_image1 = target_dict['target_img'][k].numpy()
            new_image1 = new_image1*255 / new_image1.max()
            new_image1 = np.uint8(new_image1)
            new_image1 = new_image1.transpose(1, 2, 0)
            # cv2.imshow('222', new_image1)
            # cv2.waitKey(5000)
            # cv2.destroyAllWindows()
            for i, x_y in enumerate(xys1):
                print(111)
                cv2.circle(new_image1, (int(x_y[0]), int(x_y[1])), 2, (0, 0, 255), -1)

说明：
target_dict[‘target_img’][k] 是（CHW）的 tensor，其数值范围是【0,1】类型是【torch.float32】
转换后的 new_image1 是 （HWC）的array，其数值范围是【0,255】类型是【np.uint8】

这里的new_image1直接 cv2.imshow 没有问题，但是 cv2.circle 就一直报下方的错误，排查不出原因

报错

    cv2.circle(new_image1, (int(x_y[0]), int(x_y[1])), 2, (0, 0, 255), -1)
cv2.error: OpenCV(4.6.0) :-1: error: (-5:Bad argument) in function 'circle'
> Overload resolution failed:
>  - Layout of the output array img is incompatible with cv::Mat
>  - Expected Ptr<cv::UMat> for argument 'img'
>  

解决方法

加入两行代码即可，原因不知道，但可以跑通了

关键！！！！！

            b, g, r = cv2.split(new_image1)
            new_image1 = cv2.merge([r, g, b])

        for k, xys1 in enumerate(output_reproj['keyp_2d']):
            new_image1 = target_dict['target_img'][k].numpy()
            new_image1 = new_image1*255 / new_image1.max()
            new_image1 = np.uint8(new_image1)
            new_image1 = new_image1.transpose(1, 2, 0)
            '''修改后的'''
            b, g, r = cv2.split(new_image1)
            new_image1 = cv2.merge([r, g, b])
            # cv2.imshow('222', new_image1)
            # cv2.waitKey(5000)
            # cv2.destroyAllWindows()
            for i, x_y in enumerate(xys1):
                print(111)
                cv2.circle(new_image1, (int(x_y[0]), int(x_y[1])), 2, (0, 0, 255), -1)

更简单的方法，数值范围不用转到【0,255】，保留原本的【0,1】也可以

        for k, xys1 in enumerate(output_reproj['keyp_2d']):
            new_image1 = target_dict['target_img'][k].numpy()
            new_image1 = new_image1.transpose(1, 2, 0)
            b, g, r = cv2.split(new_image1)
            new_image1 = cv2.merge([r, g, b])
            # cv2.imshow('222', new_image1)
            # cv2.waitKey(5000)
            # cv2.destroyAllWindows()
            for i, x_y in enumerate(xys1):
                print(111)
                cv2.circle(new_image1, (int(x_y[0]), int(x_y[1])), 2, (0, 0, 255), -1)

参考链接：CV2,Image,PIL图片读取，并且与Tensor相互转化并展示

吐槽一下

图像可视化还是 import matplotlib.pyplot as plt 用着舒服，tensor 和 array 都兼容，单通道和3通道也都可以，但是在图上画关键点还是用 import cv2，plt 都没这些功能吗，很烦！