saranshkarira / CNN-Visual-Interpretor-PyTorch

Implementations of various CNN visualization Techniques in PyTorch

ConvNets Visualizer in PyTorch

Implementations of various CNN Visualization Techniques in PyTorch. These notebooks are implemented in Google Colaboratory.

Requirements:

You will need a Google Colaboratory Account, all rest dependencies are satisfied within the notebook itself. Note : Select Runtime Environment as Python3 and Hardware as GPU in Colaboratory

Implemented:

• Deep Dream
• Layer Activations
• Per Filter Activations
• Weight/Feature Visualization
• Occlusion
• Saliency : Vanilla Backprop
• Saliency : Guided Backprop
• Smooth Grad
• Neural Texture Synthesis
• Neural Style Transfer
• Semantic Dictionaries
• GradCam
• Gradient Ascent

Deep Dream (Wait for gif to load)

Total Activations of Each Layer

Per Filter Activations of a Selected Layer

Filters of a Selected Layer(Conv1)

HeatMap by Occlusion

Saliency by Vanilla Backprop

Saliency by Guided Backprop

Saliency by Smooth Grad