namikosaito / cae_vae_prediction_sample Goto Github PK

Image reconstruction with 4 dimentional detection
CAE, VAE with additional layer for deprediction, connecting to the latent (middle) layer.



0. Put dataset in data directry with separate files.

1. Data processing
    - run data_processing.py
    - change plenty of png data to npy file, which helps calculation cost smaller
    - normalized [0, 255] -> [0.0, 1.0]

2. train CAE, VAE
    - run CAE(or VAE)_learning.py
    - save best model, whose error is the smallest

3. test CAE, VAE
    - run CAE(or VAE)_test.py
    - output reconstruction image data
    - extract latent space value
    - output prediction result

4. PCA on latent space value
    - use PCA.py
    - output analysis with some graphs