The solution is not complete:

python3 ./repro.py 
Traceback (most recent call last):
  File "./repro.py", line 21, in <module>
    imagePoints, _ = cv2.projectPoints(objectPoints, rvec, tvec, cameraMatrix, distCoeffs)
cv2.error: OpenCV(4.10.0-dev) /home/alexander/Projects/OpenCV/opencv-master/modules/calib3d/src/calibration.cpp:613: error: (-5:Bad argument) Translation vector must be 1x3 or 3x1 floating-point vector in function 'cvProjectPoints2Internal'

The similar should be done for tvec.

@asmorkalov could you show which data format you used as input? The translation vector should not be related to this issue. To me this looks more like the type assertion in projectPoints reacts to some invalid input

@asmorkalov could you show which data format you used as input? The translation vector should not be related to this issue. To me this looks more like the type assertion in projectPoints reacts to some invalid input

opencv/modules/calib3d/src/calibration.cpp

Lines 621 to 625 in b8ec949

	if( (CV_MAT_DEPTH(t_vec->type) != CV_64F && CV_MAT_DEPTH(t_vec->type) != CV_32F) ||
	(t_vec->rows != 1 && t_vec->cols != 1) ||
	t_vec->rows*t_vec->cols*CV_MAT_CN(t_vec->type) != 3 )
	CV_Error( cv::Error::StsBadArg,
	"Translation vector must be 1x3 or 3x1 floating-point vector" );

I realized just now that the original issue python code uses invalid data input. projectPoints expects tvec to be a 3x1 or 1x3 vector, not a 3x3 matrix.
The C++ example in the comments uses the right format.
#25318 (comment)

@j3knk, thank you for the patch! Could you please add a test that reproduces error in the original code?

@vpisarev I have used this example code from #25318 (comment) for testing. Did not have the time to create some randomized tests yet, but this triggers the original bug for sure.

#include <iostream>
#include "opencv2/core.hpp"
#include "opencv2/calib3d.hpp"

int main( int argc, char* argv[] )
{
    using type = float;
    cv::Mat objectPoints = (cv::Mat_<type>(3,3) << 181.24588 ,  87.80361 ,  11.421074,
                                                    87.17948 , 184.75563 ,  37.223446,
                                                    22.558456,  45.495266, 246.05797);
    cv::Mat rvec = (cv::Mat_<type>(3,3) << 0.9357548 , -0.28316498,  0.21019171,
                                            0.30293274,  0.9505806 , -0.06803132,
                                            -0.18054008,  0.12733458,  0.9752903);
    cv::Mat tvec = (cv::Mat_<type>(3,1) << 69.32692 ,  17.602057, 135.77672);
    cv::Mat cameraMatrix = (cv::Mat_<type>(3,3) << 214.0047  ,  26.98735 , 253.37799,
                                                    189.8172  ,  10.038101,  18.862494,
                                                    114.07123 , 200.87277 , 194.56332);

    cv::Mat distCoeffs = cv::Mat_<type>(4, 1, (type)0);

    cv::Mat out;
    cv::projectPoints(objectPoints, rvec, tvec, cameraMatrix, distCoeffs, out);

    std::cout << out << std::endl;
    return 0;
}

Equivalent in Python:

import cv2
import numpy as np

objectPoints = np.array([[181.24588 ,  87.80361 ,  11.421074],
       [ 87.17948 , 184.75563 ,  37.223446],
       [ 22.558456,  45.495266, 246.05797 ]], dtype=np.float32)
rvec = np.array([[ 0.9357548 , -0.28316498,  0.21019171],
       [ 0.30293274,  0.9505806 , -0.06803132],
       [-0.18054008,  0.12733458,  0.9752903 ]], dtype=np.float32)
tvec = np.array([ 69.32692 ,  17.602057, 135.77672 ], dtype=np.float32)
cameraMatrix = np.array([[214.0047  ,  26.98735 , 253.37799 ],
       [189.8172  ,  10.038101,  18.862494],
       [114.07123 , 200.87277 , 194.56332 ]], dtype=np.float32)
distCoeffs = distCoeffs = np.zeros((4, 1), dtype=np.float32)

imagePoints, _ = cv2.projectPoints(objectPoints, rvec, tvec, cameraMatrix, distCoeffs)
print(imagePoints)