Hi there,

We recently released a new mini-course inside PyImageSearch University called HydraNets & Multi-Task Learning for Autonomous Vehicles with PyTorch!

And we are super excited about this course.

So today, we present the course author Jeremy Cohen to tell you more about what you'll learn inside this course.

Enters Jeremy:

Hello PyImageSearch Member!

I'm Jeremy, and I run Think Autonomous, a cutting-edge course platform that teaches engineers how to build advanced applications of AI such as Self-Driving Cars. 

My main goal is to help you work with futuristic companies and have fun in your job by sending you daily emails that teach you how to become a cutting-edge engineer.

Today, I will share a few reasons why you should learn about HydraNets.

I strongly believe that Computer Vision isn't just about canny-edge detection and OpenCV.

Most of what you'll find online is about the fundamentals of Computer Vision. And to me, PyImageSearch University belongs to the best places to learn Computer Vision and Deep Learning. 

I don't know about you, but when I'm on my own, it can quickly become harder than I expected.

For example, as a self-driving car engineer, I once had to run semantic segmentation and object detection in parallel, which wouldn't work. I could run one algorithm, but not the two. The computer would crash.

This was a problem, and I had 7 or 8 more models to stack.

This is where I realized that companies producing self-driving cars were using a Multi-Task Learning architecture, capable of adding heads to a model. Each head would be responsible for a task.

In the new HydraNets mini-course, you will learn how to build semantic segmentation and monocular depth estimation architectures for self-driving cars in a single model.

This is an advanced concept and not open to everybody.

So, to qualify, you need to be able to 

  • code in Python
  • have the fundamentals of Computer Vision
  • know about CNNs and Deep Learning

If this is your case, here are a few things you'll learn in this course:

  • How to Build Advanced Multi-Task Models with PyTorch (even if you're a complete beginner) 
  • 2 Beheading Techniques to add heads to any pretrained Neural Network
  • PyTorch 101 — My cookbook to build a DataLoader, Create Models, and Train Parameters from scratch with PyTorch
  • How to tune your hyperparameters when heads are learning at different rates
  • Multi-Task Learning in Computer Vision — Deep Dive inside an untold Experimentation that reveals which task you should train together ... and avoid mixing!
  • The Intermediate Python Concepts you should know to make your code look more professional
  • ✅ PROJECT: Build your first multi-task learning algorithm from scratch with PyTorch to process images and do binary classification, regression, and multi-class classification
  • The Encoder-Decoder Architectures used to build multi-head decoders that work better and faster than single task networks
  • ✅ HYDRANET PROJECT — Build a HydraNet trained with 2 to 3 heads that does real-time semantic segmentation and depth estimation in self-driving cars

After this course, you'll be proficient in PyTorch, and you'll know how to create advanced projects such as this from one of our members that combines the 2 outputs into a 3D point cloud:

When I created this course, I tried to make it as open as possible. But if you don't have the prerequisites, you can still enroll in PyImageSearch University, build these prerequisites, and then follow the course.

Anyway, this is what I wanted to share with you. This topic had been incredibly popular with my cutting-edge students, and some of them already found jobs just by sharing their results online.

I'm super excited to bring these skills to you in this mini-course, and I'm convinced that it will help you build advanced skills that will make your profile rare and valuable.

To enroll, here's the button to the University:

Start the course


Jeremy & the PyImageSearch Team