NVIDIA Announces Open-Source CV-CUDA Project
When offered to be briefed in advance on a new "open-source" CUDA project my interest certainly piqued... Yes, NVIDIA has a new project that is open-source but still reliant on CUDA with NVIDIA's proprietary APIs and closed-source software/driver implementation. Just like many of NVIDIA's other existing open-source projects designed for accelerated GPU computing. In any case, this new project is CV-CUDA and it's an open-source project designed to handle image pre- and post-processing for accelerating computer vision (CV) workloads.
The CV-CUDA open-source library is designed to help build accelerated end-to-end computer vision and image processing pipelines. CV-CUDA initially consists of 50+ CV algorithms and supports integration of custom kernels, zero-copy interfaces, and other modern features.
NVIDIA claims CV-CUDA can process 10x as many streams on a single GPU. CV-CUDA can interface with C/C++ and Python applications as well as integrating into existing deep learning frameworks and other software.
NVIDIA plans to release CV-CUDA in early access form starting in December while a beta is planned for next March.