在c ++中的opencv convertTo和Python中的手动转换之间有不同的结果

Question

I'm trying to port a code from c++ to python, where at some point a frame is extracted from a .oni recording (OpenNI2), scaled to 8 bit and saved as jpg. I use OpenCV function convertTo in c++, which is not available in python, so reading the documentation I'm triying to do the same operation manually, but something is wrong.

This is the c++

cv::Mat depthImage8;

double maxVal = 650.0;
double minVal = 520.0;
depthImage.convertTo(depthImage8, CV_8UC1,  255.0/(maxVal - minVal), -minVal * 255.0/(maxVal - minVal));
cv::imwrite(dst_folder + "/" + std::to_string(DepthFrameIndex) + "_8bit.jpg", depthImage8);

which produce:

This is the Python version:

depth_scale_factor = 255.0 / (650.0-520.0)
depth_scale_beta_factor = -520.0*255.0/(650.0-520.0)
depth_uint8 = (depth_array*depth_scale_factor+depth_scale_beta_factor).astype('uint8')

which produce:

This code seems to work, but however images generated are different, while the original one (16UC1) are identical (already checked and they match pixel by pixel), so there should be something wrong in the conversion functions.

Answer 1

Thanks to the comments I came up with the solution. As stated by users michelson and Dan Masek Opencv performs saturate_cast operation, while numpy don't. So in order to get the same result, Python version must be:

    depth_uint8 = depth_array*depth_scale_factor+depth_scale_beta_factor
    depth_uint8[depth_uint8>255] = 255
    depth_uint8[depth_uint8<0] = 0
    depth_uint8 = depth_uint8.astype('uint8')