Library: MIR Eval library ported to pytorch

less than 1 minute read

Published:

I’ve ported MIR eval library to PyTorch enabling a faster evaluation.
It’s is available here –> torch_mir_eval

  • Only the Source separation functions has been ported, even though the project is opensource and open to grow.

Benchmark

In order to check the suitability of using a GPU to carry out. To do so we are gonna use the following components

CPU: AMD Ryzen Threadripper 1920X 12-Core Processor
GPU: Quadro P6000
Driver Version: 440.100
CUDA Version: 10.2
Pytorch: 1.6.0
numpy: 1.16.4

TODO