opencv: BUG:TypeError: Expected Ptr for argument 'img'

OpenCV =>3.10.37
Operating System / Platform => Windows 64 Bit -python

import cv2
import sys
import numpy as np
import datetime
import os
import glob

import time



img = cv2.imread('t1.jpg')

# print(type(img))
img=np.rot90(img,-1)



# (h, w) = img.shape[:2]  # 10
# center = (h // 2, w // 2)  # 11
# M = cv2.getRotationMatrix2D(center, 90, 1.0)  # 15
# img = cv2.warpAffine(img, M, (w, h))  # 16

color = (0, 0, 255)
print(img)

cv2.rectangle(img, (10,20 ),(30,40) ,color,2)


cv2.imwrite("5_flip.jpg",img)

When I ran this script, I encountered the following error：TypeError: Expected Ptrcv::UMat for argument ‘img’

About this issue

Original URL
State: closed
Created 4 years ago
Comments: 18 (1 by maintainers)

Most upvoted comments

Somehow img = np.array(img) still doesn’t work for me but img = img.copy() works.

+36

louis925 on Nov 15, 2020

Just had this issue and resolved it by the explicit img = np.array(img) right before the operation that fails. It somehow helps even though type(image_db) returns <class 'numpy.ndarray'> both before and after this line.

+29

yurkovak on Aug 18, 2020

@HanwenCao In my case the issue was in what was discussed in #15895. The bug appeared after flipping RGB to BGR with img = img[:, :, ::-1]. After replacing this line with the corresponding opencv operation like this img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB) I was able to get rid of img = np.array(img).
Please refer to this comment for a better explanation of this behavior. Unfortunately, I wasn’t able to find what exactly “numpy on-the-fly reshaped array” means in a broader sense.

Just for the record:

OpenCV: 4.2.0
Python: 3.7.0
OS: 16.04

+25

yurkovak on Sep 7, 2020

change this " img = cv2.imread(‘t1.jpg’) " to " ret, img = cv2.imread(‘t1.jpg’) " may slove your problem.

+22

SCZeiDan on Sep 21, 2020

Try to investigate using #15918

img=np.rot90(img,-1)

OpenCV can’t handle views as InputOutputArrays (there are several requirements for conversion).

Try to make regular numpy array:

img=np.array(np.rot90(img,-1))

+10

alalek on Aug 18, 2020

np.ascontiguousarray also help mitigate this issue.

huntzhan on Dec 20, 2020

same as @louis925 , none of the other solutions worked for me but img = img.copy() did work. No idea why, because the dtype of img for me was already uint8 and the type was numpy.ndarray.

The command I was trying to run was

cv2.normalize(img, img, 0, 255, cv2.NORM_MINMAX, cv2.CV_8U)

Note that changing the dst argument to None also fixed this issue. Very curious as to what’s causing this problem.

ash-jha on Nov 20, 2020

When using cv2.circle I had to convert my 2D array using: img = np.array(img, dtype=np.uint8 to avoid this error.

saftanas on Jul 27, 2022

In my case, using im = cv2.cvtColor(im, cv2.COLOR_BGR2RGB) after np.rot90(img,-1) works. However, just using img=np.array(np.rot90(img,-1)) doesn’t work.

Great help. img = np.array(img) and img=img[:,:,::-1] both doesn’t work. But img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB) worked.

www516717402 on Nov 12, 2020

You can use np.ascontiguousarray(img), for example in your case: cv2.rectangle(np.ascontiguousarray(img), (10,20 ),(30,40) ,color,2)

navidnadery on Jul 19, 2021

In my case, using im = cv2.cvtColor(im, cv2.COLOR_BGR2RGB) after np.rot90(img,-1) works. However, just using img=np.array(np.rot90(img,-1)) doesn’t work.

sidml on Nov 2, 2020