CUDA Image & Video Resizer for NVIDIA GPU
Updated:
Fastvideo's CUDA Resizer scales grayscale and colour images and video on the NVIDIA GPU with high quality and very high speed. It uses a Lanczos kernel with floating-point precision and is a module of our CUDA Image & Video Processing SDK, so resize can sit anywhere in a real-time pipeline.
Features
- Input: 8/16-bit per channel RGB, PGM, PPM, BMP, JPEG, or a byte array in CPU/GPU memory
- Upscale or downscale to any size, quickly and with high quality
- Lanczos resize algorithm, floating-point internal calculations
- Output: JPEG, BMP, PPM, PGM, DIB
- Windows 10/11, Linux Ubuntu, Linux4Tegra (L4T); latest NVIDIA GPUs
Performance
High-quality resize of a colour Full HD frame (1920×1080, 24-bit) down to 960×576 runs at about 3000 fps on a GeForce GTX 1080. For best results, combine the resizer with the other SDK modules in a single GPU pass.
Roadmap
- CUDA Resizer integrated into Fast CinemaDNG Processor – done
- Lower GPU memory usage, more input/output formats and a new high-quality resize algorithm