• Optimize --vpp-kfm / --vpp-degrain.
• Fix libvmaf CUDA implementation and access violation. ( #788 )
• Update Linux package CUDA to 11.8.
• Optimize --vpp-onnx. (Thanks @teaching-droid!)
  • Better performance by adding with zero-copy path.
  • Support frames setting of registered models.
  • Better model handling.