Learn from the best in the field with our exclusive computer vision speaker series. [Register now](https://www.nvidia.com/en-us/events/computer-vision-speaker-series/?nvid=nv-int-unbr-911558) for free! 

# CV-CUDA

CV-CUDA™ is an open-source library that enables building high-performance, GPU-accelerated pre- and post-processing for [AI computer vision](/computer-vision) applications in the cloud at reduced cost and energy.

  

[Download on GitHub](https://github.com/CVCUDA/CV-CUDA)      [Share Your Use Case](/cv-cuda/early-access)

* * *

 ![AI Computer Vision](https://d29g4g2dyqv443.cloudfront.net/sites/default/files/akamai/cuda/cv-cuda/cv-cuda-850x480.jpg "AI Computer Vision")  

## Use Cases

Explore common use cases with AI imaging and computer vision workloads deployed at scale in the cloud.

### Image Understanding

Image understanding involves AI algorithms interpreting and processing visual data to recognize patterns, objects, and context, paving the way for applications like facial recognition, medical imaging, and scene understanding.

  
[Learn More](https://catalog.ngc.nvidia.com/orgs/nvidia/teams/tao/collections/tao_computervision)

### Generative AI

Generative AI algorithms produce new content or data, imitating the patterns they learn from existing data, enabling tas  
ks like image creation, text generation, and style transfer.

  
[Learn More](https://www.nvidia.com/en-us/research/ai-playground/)

### 3D Worlds

3D worlds are digital environments that represent space in three dimensions, offering immersive experiences for users in gaming, simulations, and virtual reality platforms.

  
[Learn More](https://www.nvidia.com/en-us/omniverse/)

### HD Mapping

HD mapping creates highly detailed digital representations of the physical world, essential for the precision and decision-making processes of autonomous vehicles.

  
[Learn More](https://www.nvidia.com/en-us/self-driving-cars/hd-mapping/)

* * *

## Key Features

  

CV-CUDA provides a specialized set of 45+ highly performant computer vision and image processing operators. CV-CUDA also offers:

- C, C++, and Python APIs
- Batching support, with variable shape images
- Zero-copy interfaces to deep learning frameworks like [PyTorch](https://pytorch.org/) and [TensorFlow](https://www.tensorflow.org/)
- An NVIDIA Triton™ Inference Server example using CV-CUDA and NVIDIA® [TensorRT™](https://developer.nvidia.com/tensorrt)
- End-to-end GPU-accelerated object detection, segmentation, and classification examples

  

View a full list of the operators in the [CV-CUDA documentation](https://cvcuda.github.io/).

 ![CV-CUDA Library](https://developer.download.nvidia.com/images/cv-cuda-stack-630x600.jpg "CV-CUDA Library")

* * *

## CV-CUDA Benefits

 ![Computer Vision Cloud Applications](https://d29g4g2dyqv443.cloudfront.net/sites/default/files/akamai/cuda/cv-cuda/m48-edge-computing(1).svg "Computer Vision Cloud Applications")
### Specialized Set of Kernels for Cloud-Based Use Cases

 ![Computer Vision Operators](https://d29g4g2dyqv443.cloudfront.net/sites/default/files/akamai/cuda/cv-cuda/m48-workflow-complex.svg "Computer Vision Operators")
### Efficient, Hand-Optimized Kernels That Save Cost and Energy

 ![CV-CUDA Integration](https://d29g4g2dyqv443.cloudfront.net/sites/default/files/akamai/cuda/cv-cuda/m48-fleet-command.svg "CV-CUDA Integration")
### Lightweight and Flexible for Integrating Into Frameworks

* * *

## Up to 49X End-to-End Throughput Improvement

CV-CUDA lets you move your bottlenecked pre- and post-processing pipelines from the CPU to the GPU, boosting throughput for complex workflows.

  

For a typical video segmentation pipeline, CV-CUDA enabled an end-to-end 49X speedup using NVIDIA [L4 Tensor Core GPUs](https://www.nvidia.com/en-us/data-center/l4/). With the latest and most efficient NVIDIA GPUs and CV-CUDA, developers of cloud-scale applications can save tens to hundreds of millions in compute costs and eliminate thousands of tons in carbon emissions.

### Video Segmentation Pipeline (End-to-End) 

1080p, 30fps

* * *

## Interoperability

CV-CUDA is interoperable with the following libraries, SDKs, and frameworks.

- [nvJPEG libraries](https://developer.nvidia.com/nvjpeg)
- [Video Codec](https://developer.nvidia.com/video-codec-sdk)
- [Video Processing Framework (VPF)](https://github.com/NVIDIA/VideoProcessingFramework)
- [TAO Toolkit](https://developer.nvidia.com/tao-toolkit)

- [TensorRT](https://developer.nvidia.com/tensorrt#resources)
- [Triton Inference Server](https://developer.nvidia.com/triton-inference-server)
- [PyTorch](https://pytorch.org/)
- [TensorFlow](https://www.tensorflow.org/)

  
  
  

## Global Industry Adoption

From content understanding to visual search and generative AI, customers are adopting CV-CUDA for their AI computer vision use cases.

 ![NVIDIA Partner for CV-CUDA](https://d29g4g2dyqv443.cloudfront.net/sites/default/files/akamai/cuda/cv-cuda/baidu-logo.svg "NVIDIA Partner for CV-CUDA")

 ![Microsoft- NVIDIA Partner for CV-CUDA](https://d29g4g2dyqv443.cloudfront.net/sites/default/files/akamai/cuda/cv-cuda/logo-bing.svg "NVIDIA Partner for CV-CUDA")

 ![Runway- NVIDIA Partner for CV-CUDA](https://d29g4g2dyqv443.cloudfront.net/sites/default/files/akamai/cuda/cv-cuda/runway-logo(2).svg "NVIDIA Partner for CV-CUDA")

 ![NVIDIA Partner for CV-CUDA](https://d29g4g2dyqv443.cloudfront.net/sites/default/files/akamai/cuda/cv-cuda/tencent-cloud-logo.svg "NVIDIA Partner for CV-CUDA")

* * *

## In the News

 ![Computer Vision Segmentation Pipeline](https://d29g4g2dyqv443.cloudfront.net/sites/default/files/akamai/cuda/cv-cuda/reducing-cost-for-ai-based-computer-vision-630x354.jpg "Computer Vision Segmentation Pipeline")

### Increasing Throughput and Reducing Cost for AI-Based Computer Vision With CV-CUDA

CV-CUDA enables real-time, high-performance cloud-scale applications with demands for lower latency and higher throughput.

  

[Read how CV-CUDA increases throughput while also reducing both cost and energy consumption](/blog/increasing-throughput-and-reducing-costs-for-computer-vision-with-cv-cuda/)

 ![CV-CUDA for Visual Search](https://d29g4g2dyqv443.cloudfront.net/sites/default/files/akamai/cuda/cv-cuda/adopting-cv-cuda-630x354.jpg "CV-CUDA for Visual Search")

### NVIDIA Announces Microsoft, Tencent, Baidu Adopting CV-CUDA for Computer Vision AI

CV-CUDA is helping customers build and scale AI-based imaging and computer vision pipelines.

  
  
  

[Read how industry leaders are adopting CV-CUDA](https://blogs.nvidia.com/blog/2023/03/21/cv-cuda-ai-computer-vision/)

 ![CV-CUDA Video Application](https://d29g4g2dyqv443.cloudfront.net/sites/default/files/akamai/cuda/cv-cuda/computer-vision-cloud-applications-630x354.jpg "CV-CUDA Video Application")

### NVIDIA Introduces Open-Source Project to Accelerate Computer Vision Cloud Applications

CV-CUDA combines accelerated image pre- and post-processing algorithms and tools to process higher image throughput and lower cloud computing cost.

  

[Read how CV-CUDA can accelerate pre- and post-processing pipelines](https://blogs.nvidia.com/blog/2022/09/20/computer-vision-cloud/)

* * *

## Videos and Webinars

## Additional Resources

- [Watch Webinar: Overcoming Pre- and Post-Processing Bottlenecks in AI Computer Vision Pipelines](https://www.nvidia.com/en-us/on-demand/session/gtcspring23-s51182/)(42:27 Minutes)
- [See the Difference CV-CUDA Makes: Runway Optimizes AI Image and Video Generation Tools With CV-CUDA](https://www.youtube.com/watch?v=cxS4pT16_XQ)(01:18 Minutes)
- [Review Documentation: CV-CUDA Developer Guide](https://github.com/CVCUDA/CV-CUDA/blob/release_v0.2.x/DEVELOPER_GUIDE.md)

Help make CV-CUDA better by sharing your feedback and how you’re using it. We may follow up with you to continue the conversation.

  
[Share Your Use Case](/cv-cuda/early-access)  

You must be a member of the [NVIDIA Developer Program](/developer-program) and be logged in with your organization’s email address. We don’t accept applications from personal email accounts.