CV-CUDA

Why we included this project

For teams whose vision pipelines stall on CPU-side preprocessing, CV-CUDA moves that work onto the GPU. It provides a library of accelerated operators for common transforms like resizing and color conversion, so the whole pre- and post-processing path stays on the device instead of bouncing data back to the host. The library plugs into existing C/C++ and Python AI frameworks, and it pairs with NVIDIA codecs for decoding images and video, which makes it a natural fit for throughput-heavy jobs like feeding large batches through an inference service. If your current path uses OpenCV on CPU, this is worth benchmarking against your own workload, especially if you already work in the CUDA ecosystem and want a straightforward Python API without dropping to low-level code.