DNN-based_source_separation icon indicating copy to clipboard operation
DNN-based_source_separation copied to clipboard

Published 20 hours ago • verified publisher icon tky823

Implementation of D3Net

Open tky823 opened this issue 4 years ago • 11 comments
trafficstars

Reference: "D3Net: Densely connected multidilated DenseNet for music source separation"

tky823 avatar Mar 11 '21 11:03 tky823

I'm not sure of # channels before frequency concatenation. The # of channels depends on the growth rate and # of D2 blocks. I added bottleneck convolution so that both frequency bands have the same channels.

https://github.com/tky823/DNN-based_source_separation/blob/48621f1dbb015454246f9df6e87d20788fc75114/src/models/d3net.py#L107-L109

tky823 avatar Mar 13 '21 06:03 tky823

What needs to be fixed

  • [x] multi dilated convolution
    • [x] timing of batch normalization
  • [x] # of output channels of D2 block
  • [x] order of D3 block and downsampling layer in Down D3 block
  • [x] upsampling layer

tky823 avatar Jun 05 '21 08:06 tky823

Now, I updated D3Net architecture.

tky823 avatar Jun 06 '21 10:06 tky823

Hello, @tky823. I participate in the Music Demixing Challenge (4th place on leaderboard A). I suggest you write a training script for D3Net and join a team with me.

lyghter avatar Jun 27 '21 11:06 lyghter

Hi, @lyghter. I'm now writing the training code. I am not sure if it will be available soon, but I plan to add it.

tky823 avatar Jul 04 '21 14:07 tky823

The challenge will end on July 31st. If you write the training code this month, I will try to train the model and use it in my solution. Sony's nnabla implementation has too slow inference on CPU. It cannot be used in the challenge.

lyghter avatar Jul 04 '21 14:07 lyghter

I invite you to join my team and suggest you keep the new code private until the end of the challenge.

lyghter avatar Jul 04 '21 14:07 lyghter

I am currently 4th on Leaderboard A and 5th on Leaderboard B. Top-3 from A and top-3 from B will receive prizes.

lyghter avatar Jul 04 '21 15:07 lyghter

I'm not sure how my implementation of D3Net will work, so I don't know if I'll be able to participate anytime soon. If I can help, I will join your team. I work on other tasks for about a week. Maybe I will be able to join after that.

tky823 avatar Jul 06 '21 05:07 tky823

Hello @tky823 Take a look at this. It looks like this repo contains pytorch implementation of D3Net and training code. I just found it and haven't tried it yet.

lyghter avatar Jul 12 '21 12:07 lyghter

@lyghter I'm sorry I couldn't help you. Now, I'm sharing the scripts and results in egs/musdb18/d3net.

tky823 avatar Sep 03 '21 07:09 tky823