Nvidia launched the open source project CV-CUDA, which processes 10 times more streams than a single GPU.

CTOnews.com Sept. 25, Nvidia learned that to help with faster and more efficient large-scale image processing, Nvidia has launched CV-CUDA, an open source library for building accelerated end-to-end computer vision and image processing pipelines.

Nvidia says Internet traffic is dominated by video. Such videos will increasingly be enhanced by AI special effects and computer graphics. At the same time, the rapid growth of social media and video sharing services has encountered difficulties, including the rising cost of cloud computing and bottlenecks in AI-based imaging processing and computer vision pipelines.

It is reported that CV-CUDA can accelerate AI special effects, such as re-lighting, reset, blurred background and ultra-high resolution. CV-CUDA provides more than 50 high-performance computer vision algorithms, a development framework that can easily implement a custom kernel, and many zero replication interfaces.

According to Nvidia, CV-CUDA streams 10 times as much as a single GPU-based stream, helping developers accelerate video content creation, 3D worlds, image-based recommendation systems, image recognition and video conferencing.

CV-CUDA accelerates pre-and post-processing production processes through a manually optimized CUDA kernel and is natively integrated into C / C++, Python, and common deep learning frameworks such as PyTorch. CV-CUDA will become one of the core technologies to accelerate AI workflow in NVIDIA Omniverse. In addition, developers can try the code first in December, and the beta version will be released in March next year.