We use OIDN with a stream of inputs originally in half-float format (IEEE 754 half-float format, RGB/RGBA). Currently we convert the input images to 32-bit floats with SIMD intrinsics prior to denoising which works okay but does use extra memory and costs time; are there any plans to extend the RT filter to natively support half-float formats?
We use OIDN with a stream of inputs originally in half-float format (IEEE 754 half-float format, RGB/RGBA). Currently we convert the input images to 32-bit floats with SIMD intrinsics prior to denoising which works okay but does use extra memory and costs time; are there any plans to extend the RT filter to natively support half-float formats?
Thanks