v0.1.0-preAlpha

CV-CUDA-0.1.0 is the first release of CV-CUDA. This release is for evaluation purposes only.

Release Highlights

This CV-CUDA release includes the following key features:

  • Core components including Image and Tensor with Batch support

  • Utilities to help write CUDA kernels

  • 6 Operators - Reformat, Resize, Custom Crop, Normalize, PadAndStack, ConvertTo

  • Tensor interoperability with pytorch/gpu, Image interoperability with pytorch/gpu, pillow/cpu, opencv/cpu

  • Python bindings

  • Sample applications

  • API documentation

Compatibility

This section highlights the compute stack CV-CUDA has been tested on

  • Ubuntu x86 >= 20.04

  • CUDA driver >= 11.7

The Sample applications based on TensorRT have been tested with TensorRT >= 8.5

Known Issues

  • There will be few updates in the Tensor API, Image Formats and Operator names in the next release

  • Limitations in the usage of the operators which are described in the API documentation

License

Nvidia Software Evaluation License