C++ / Python inconsistency : int point coordinates instead of the original float

[Q4pLJ1yy]

Hi,
It seems DetectMarkers has not the same behavior between the original C++ and this Python wrapper : convertToTuple applies a cast to int that is not present in the C++ library :
std::vector<std::vector<float>> contours_vec; for (cv::Point2f pt : corners[i]) { contours_vec.push_back({static_cast<float>((int)pt.x), static_cast<float>((int)pt.y)}); }
Is this type cast to int necessary between float and Point2f ? The accuracy pointed out in the original paper must be downgraded by this step.

[ManfredStoiber]

Hi,
as mentioned in this issue, I honestly can't recall the exact reason why I originally clipped the values to integers. I tested it without clipping and had no problems. If you'd like, I can build and distribute a new version without the casting.

Best regards

[Q4pLJ1yy]

Hi,
That would be great, I had some issues compiling locally. I am interested too by the full resolution of the output for a Python testbed, if only to see if I can improve positioning results compared to regular subPixelCorners applied after the fact. Thanks !
Best regards