CV-CUDA Release v0.2.1

[milesp-nvidia]

CV-CUDA Release v0.2.1

General

• Added DLPack interop support to Python
• Added CV-CUDA build package for various Python versions
• Updated documentation to include NVIDIA VPF update
• Downgraded C++ version required from 20 to 17
• Made public headers compatible with C++11
• Changed default CUDA stream in Python to be 0
• Removed library build dependency on OpenSSL
• Relaxed nvcv::Tensor bit restrictions on strides
• Updated Segmentation sample app with NVIDIA VPF and Joint Bilateral Filter

New Operators

• Joint Bilateral Filter

Improved Testing

• Full operator test coverage for interleaved (HWC) and planar (CHW) tensors
• Removed Numba dependency on tests

Bug Fixes

• Addressed BUG #25 TypeError: as_tensor(): incompatible function arguments. The following argument types are supported
• Addressed BUG #26 libdl-2.17_stub.so: file format not recognized; treating as linker script
• Addressed BUG #28 cannot find -lcvcuda
• Fixed memory leak from exception thrown by constructor
• Fixed CMake code generation for header compatibility tests
• Fixed reformated from CHW to NHWC error
• Fixed NVCV_ASSERT_STATUS macro failing to compile prompt added when testing for exception
• Tensor shapes must be exported as tuples, not lists, in Python
• cudaIllegalMemoryAddress prompt added in Python tests when GPU is under load
• cudaMallocs prompt added in pipelines driven by Python incur performance penalty

---

This discussion was created from the release CV-CUDA Release v0.2.1.