System Information

latest opencv

Detailed description

In nary_eltwise_layers.cpp, there is a line dispatching by data type of output mat:

https://github.com/opencv/opencv/blob/590f150d5e032165e27d81294c9b7ac710b77f11/modules/dnn/src/layers/nary_eltwise_layers.cpp#L634C10-L634C10

In the current dnn engine, we basically use either float or float16 for input (x) and output (y) mats; thats not a problem. The problem is for constsant input noes, their type can be something else, such as CV_32S in the following model:

In this case, the input x and output y has data type CV_32F, but the constant input B has data type CV_32S. However, the dispatcher calls the template with float because of data type of y, then B values are parsed in a wrong type, leading to incorrect results.

Steps to reproduce

Run the attached model with opencv dnn and check with reference input and output.

greater.zip

Issue submission checklist

• I report the issue, it's not a question
• I checked the problem with documentation, FAQ, open issues, forum.opencv.org, Stack Overflow, etc and have not found any solution
• I updated to the latest OpenCV version and the issue is still there
• There is reproducer code and related data files (videos, images, onnx, etc)