Denoising Autoencoders

Autoencoders can be used to remove noise (denoise) from an image. But how? Train the encoder part to encode a noisy image and the decoder part to reconstruct an image that is more similar to the image without noise.

* Denoising Autoencoder Results*

Dataset

The DAE is trained on Cifar-10 dataset for about 30 epochs.

CIFAR-10 Images

Output

The VAE generated vague images on start but got better with time.

• To see the results of your training from the DAE_demo notebook - Start a new terminal and execute command:

  tensorboard --logdir=./dae
  

• To see the results of your training from the DAE_VAE_DEMO notebook - Start a new terminal and execute command:

  tensorboard --logdir=./dae_vae
  

and then go to <your-public-ip-adress>:6000 (if training on an instance) or to <localhost>:6000 (if training on local system).

DAE without Variational Space

• Noising Ratio - 0.4: 40% noisy image + 60% image

DAE outputs Noising Ratio 0.4

• Noising Ratio - 0.2: 20% noisy image + 80% image

DAE outputs Noising Ratio 0.2

• No Noise

DAE outputs no Noise

Variational DAE

• Noising Ratio - 0.4: 40% noisy image + 60% image

Variational DAE outputs Noising Ratio 0.4

• Noising Ratio - 0.2: 20% noisy image + 80% image

Variational DAE outputs Noising Ratio 0.2

• No Noise

Variational DAE outputs no Noise