-
我最近复现了一下您的代码,核心代码都一样,都是从您DCASE2020的代码copy的,然后网络结构、参数跟您设置的一致、Encoder的预训练参数也是down的您的模型,但是实验结果一直很差,Spiderscore可能才0.18左右,想请教您一下原因,真的调了好久了
-
https://github.com/MihawkHu/DCASE2020_task1/blob/44187eb3c882c8c82ea0721145cbb9ea624a3b4e/task1a/10class/fcnn/train_fcnn.py#L29
There is no such exact filename matching with 'fold1_train_full.csv'
…
-
Hi, thank you for great work and sharing code. I create environment with `DCASE2020_task1/task1a/environment.yml`. Then I try to extract features and I got this error.
```
(d20-keras) yunusemre.oz…
-
Hi, I tried your python script but can't achieve this performance.
52.8% (+-0.6%) event-based F1-score
For us, F1-score was about 30%.
What should I do to reproduce ?
-
On L36: https://github.com/MihawkHu/DCASE2020_task1/blob/master/task1a/10class/extr_feat_2020_nodelta_scaled.py#L36 why do you add 1.e-8 to the spectrogram features and then take the log of it?
Als…
-
Hi, very interesting work! Currently, I am trying to reproduce to the task1a FCNN results where you reported 76.9 % accuracy with all the augmentations. I am a bit confused about the evaluation_setup …
-
`Traceback (most recent call last):
File "task1b.py", line 692, in
sys.exit(main(sys.argv))
File "task1b.py", line 233, in main
overwrite=overwrite
File "/mnt/batch/tasks/shared/LS…
-
Can I understand this code **eval_fusion_vote_10class.py**(https://github.com/MihawkHu/DCASE2020_task1/blob/2384da9d357bc6be9648701a0ae0c4b4be285e36/task1a/10class/eval_fusion_vote_10class.py)
as an …
-
Hi,
I have two questions:
1. Is the training file wrong? 'train_csv = 'evaluation_setup/fold1_train_full.csv''
2. Are totally 9 data augmentation methods used in this project?How to used the audio_…
-
Hi,
I’d like to perform some experiments and I find this library very useful. I’d like to know how to best cite your library in case I need to?
Could you please provide a Bibtex reference?
Re…