Visualizing Filters of a CNN using TensorFlow

Offered By
Coursera Project Network
In this Guided Project, you will:

Implement gradient ascent algorithm

Visualize image features that maximally activate filters of a CNN

Clock1 hour
CloudNo download needed
VideoSplit-screen video
Comment DotsEnglish
LaptopDesktop only

In this short, 1 hour long guided project, we will use a Convolutional Neural Network - the popular VGG16 model, and we will visualize various filters from different layers of the CNN. We will do this by using gradient ascent to visualize images that maximally activate specific filters from different layers of the model. We will be using TensorFlow as our machine learning framework. The project uses the Google Colab environment which is a fantastic tool for creating and running Jupyter Notebooks in the cloud, and Colab even provides free GPUs for your notebooks. You will need prior programming experience in Python. This is a practical, hands on guided project for learners who already have theoretical understanding of Neural Networks, Convolutional Neural Networks, and optimization algorithms like gradient descent but want to understand how to use the TensorFlow to visualize various filters of a CNN. Note: This course works best for learners who are based in the North America region. We’re currently working on providing the same experience in other regions.

Skills you will develop

Deep LearningArtificial Neural NetworkConvolutional Neural NetworkMachine LearningTensorflow

Learn step-by-step

In a video that plays in a split-screen with your work area, your instructor will walk you through these steps:

  1. Introduction

  2. Downloading the Model

  3. Get Submodels

  4. Image Visualization

  5. Training Loop

  6. Final Results

How Guided Projects work

Your workspace is a cloud desktop right in your browser, no download required

In a split-screen video, your instructor guides you step-by-step

Frequently asked questions

Frequently Asked Questions

More questions? Visit the Learner Help Center.