Convolutional Autoencoders with Symmetric Skip Connections

Fully convolutional networks for autoencoders with very deep connections are succeptible two things:

• Vanishing Gradients
• Significant amount of corruption of image details

To over come this paper, Image Restoration Using Convolutional Auto-encoders with Symmetric Skip Connections used skip connections between the mirrored convolutional and deconvolutional layers. Relying on this we can overcome the above mentioned issues.

An example of a building block in the proposed framework.

Dataset

The AE is trained on Cifar-10 dataset for about 100 epochs.

CIFAR-10 Images

Output

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

tensorboard --logdir=./ae_ssc

and then go to <your-public-ip-adress>:6006 (if training on an instance) or to <localhost>:6006 (if training on local system).

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

Outputs Noising Ratio 0.4

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

Outputs Noising Ratio 0.2

• No Noise

Outputs no Noise

Issue Encountered

The AE is strangely getting stuck with blue tint in case of no noise input.