Opencv build error C2666

Sangonomiya_Kokomi · July 19, 2023, 7:15am

CUDA 12.2
cudnn 8.9.2
CMake version 3.26.4
Visual Studio Enterprise 2022 versioin 17.6.4

D:\\opencv-4.8.0\modules\dnn\src\layers\../cuda4dnn/primitives/normalize_bbox.hpp(114,17): error C2666: “operator !=”: overloaded functions with similar conversion (Compile source files D:\\opencv-4.8.0\modules\dnn\src\layers\normalize_bbox_layer.cpp)
D:\\opencv-4.8.0\modules\dnn\src\layers\../cuda4dnn/primitives/region.hpp(124,17): error C2666: "operator >": overloaded functions with similar conversion (Compile source files D:\\opencv-4.8.0\modules\dnn\src\layers\region_layer.cpp)

build opencv_world.vcxproj - failed

cudawarped · July 19, 2023, 8:21am

See Compiling Opencv Error - #3 by cudawarped

and

github.com/opencv/opencv

Comment by cudawarped to Solve error: ambiguous overload for ‘operator!=’ (operand types are ‘__half’ and ‘double’)

opencv:4.x ← LongerVision:4.x

> Perhaps original issue is no caused by missing type case, but by ambiguous con…stant type. Try to change to `1.f` and `0.f`. The static cast fixes the issue with CUDA 12.2 but due to the ambiguous constant type you mentioned causes an issue with earlier versions of CUDA. The combination of both, i.e. ``` if (weight != static_cast<T>(1.0f)) if(nms_iou_threshold > static_cast<T>(0.0f)) ``` or double ``` if (weight != static_cast<T>(1.0)) if(nms_iou_threshold > static_cast<T>(0.0)) ``` works on Windows with Visual Studio 2022 for CUDA 12.1 (old math api) and CUDA 12.2.

for the fix. Unfortunately the person who initiated the PR fixing this issue has disapeared.