LUT cube size, HSV and RGB, trilinear interpolation
Updated:
A LUT is a table, and its size and colour representation decide both the result and the time: a 65 by 65 by 65 cube is not the same work as 17 by 17 by 17, and HSV is not the same as RGB. This page measures that. Fastvideo's CUDA kernels apply 2.5D and 3D cube LUTs on the NVIDIA GPU with sub-millisecond timing, so colour grading with user-defined LUTs runs comfortably in real time as part of a GPU pipeline.
Features
- Input up to 16-bit, arbitrary width and height
- 2.5D and 3D cube LUT formats; RGB and HSV colour representation
- Internal colour cube resolution up to 65×65×65; fast trilinear interpolation
- Module of the Fastvideo Image & Video Processing SDK
Performance (16-bit RGB, trilinear, GeForce GTX 1080)
All data in GPU memory, timing is GPU computation only, for a 2.5K (2432×1366) and a 4K (4032×2192) image:
| LUT | 2.5K | 4K |
|---|---|---|
| 2.5D LUT (HSV, 90×30) | 0.26 ms | 0.64 ms |
| 2.5D LUT (HSV, 90×117) | 0.26 ms | 0.65 ms |
| 3D LUT (HSV, 36×8×8) | 0.29 ms | 0.65 ms |
| 3D LUT (HSV, 36×29×16) | 0.31 ms | 0.68 ms |
| 3D LUT (HSV, 36×57×61) | 0.44 ms | 0.77 ms |
| 3D LUT (RGB, 17×17×17) | 0.22 ms | 0.55 ms |
| 3D LUT (RGB, 33×33×33) | 0.22 ms | 0.56 ms |
| 3D LUT (RGB, 65×65×65) | 0.30 ms | 0.60 ms |
The 3D LUT transform is a module of our CUDA Image & Video Processing SDK for real-time colour grading with user-defined LUTs.