DNN_TARGET_OPENCL_FP16 inference is currently available only with InferenceEngine (Intel hardware). However T8xx Mali GPUs (and many other GPUs), present in the quite popular ARM64 RK3399, are able to perform fp16 operations. This would result in a x4 inference speed up compared to fp32 inference.

I hope this could trigger a discussion on the changes involved. (Unfortunately I cannot help if OpenCL kernels need changes)

Issue submission checklist

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