shinmura0 / dcase2020_task2_solution-anomaly_detection-

DCASE2020 Task2 Self-supervised learning solution

DCASE2020 Task2(anomaly detection) 15th Solution

This is DCASE2020 Task2 Self-Supervised Learning solution (technical report).
The competition rules is here. This competition is anomaly detection problem in machine sound.

Proposed method

• Architecture

• Generate anomaly sound

• Result(ROC-AUC)

• Visualization by Faster-Grad-CAM

In preparation

• Jupyter nootbook: coming soon... complete
• Pre-trained model: comming soon... complete
• Visualization by Grad-CAM: comming soon... complete
• arxiv paper: comming soon...

dcase2020_task2_solution-anomaly_detection-'s Issues

Where can we find the code related to the pre-training of MobileNetV2 ?

For reproducibility issues, does anyone know where can I find the code used to pre-train the MobileNetV2 model with the Softmax Cross Entropy Classification and MixUp Augmentation ?
Thanks in advance!

Details of the virtual env

Please provide the details of the version of the packages used to run the notebook.

in: model_pre = model_from_json(open(model_path + 'model.json').read())
I get the following error "ValueError: Unknown layer: ReLU"

I use Keras 2.2.5 and tensorflow 1.15.3

Please provide your versions, not only keras, tensorflow, but also others e.g. a requirements.txt

Will you share your train code?

@shinmura0 Hi,Will you share your train code?

how to get the values of max_,min_,mean and sigma of the dataset?

@shinmura0 Hi. how to get the values of max_,min_,mean and sigma of the dataset?
max_ = np.max(x_train)=0.18673706
min_ = np.min(x_train)=-0.18011475
But they are different from yours(max_ = 15.6,min_ = -74.7)
Also how to get mean and sigma ?
Thanks!

Complete code

Can I upload the complete code? How does the pre training model work