How to do yolo inference with 16-bit half precision floats instead of 32-bit float?

[prnvjb]

Hi guys,

What is the key parameter that I should change in yolo v3 model/cfg or in libdarknet for inferencing with float16? I heard it increases inference speed.
would you tell me what are the other parameter to consider for inferencing yolo v3?

Thanks!!

[naturalistic]

Hi, I believe this was added in a fork located here:

The developer mentioned it increases detection 3x ect.

https://github.com/AlexeyAB/darknet

[czhaneva]

Hi, I believe this was added in a fork located here:

The developer mentioned it increases detection 3x ect.

https://github.com/AlexeyAB/darknet

Hello, would you please tell me where should I to find the implementation.
I can't find relative information.