Hi,

I love playing Rocket League in my free time.  It's a wacky game where you play soccer with rocket-powered cars. I've always found it a great way to relax, be it alone or with friends. 

Unfortunately, it does have a steep learning curve. Once I started playing in competitive mode, I got matched against players who played like their lives depended on it. I wasn't really good at it, and people started scoring on me left and right, with every trick available in their book. Quite a bit like this:


Shaolin Soccer

Honestly, it became annoying since it was one of my favorite ways of relaxing. Instead, it left me more stressed out. 

So what did I do?

Shaking off my lone wolf ideology, I started making friends who knew the game. Queuing with them not only saved me from getting constantly humiliated, but I started picking up the game surprisingly fast. I guess realizing that someone has your back alleviated a lot of pressure. 

I'd say that today's blog carries a similar message. Today's machine learning community has several tools to make your journey easier, especially in the starting phase. Using these tools can make your learning experience much easier than trying to figure out everything from scratch. One such tool is Torch Hub and its immaculate state-of-the-art (SOTA) pretrained models. 

The big picture: Torch Hub hosts a commendable collection of pretrained SOTA models for various tasks, ranging from Audio and Generative to Natural Language Processing. 

These models have been trained on standard datasets and can help you tremendously when you don't have the time or computational power to train your own models.

How it works: As explained in the last tutorial, you can host a model on PyTorch Hub by including a hubconf.py file containing an entry point inside your model repository. These SOTA models have been made available similarly, but with the added option of loading your model with pretrained weights. 

Since some tasks are more common than others, PyTorch Hub selected a few SOTA models with a code example to display in their gallery. 

In today's blog, you'll learn:

  • Basics about VGG and ResNets (the most-used models for image feature extraction)
  • Calling these models from Torch Hub and tinkering with their layers
  • Fine-tuning the models based on a simple classification problem

My thoughts: These models can make your life 10x easier. The hassle of training big architectures from scratch is absolutely nullified. Indeed, it saves you a ton of time, but the real plus point is understanding how good these models become when it comes to feature extraction. A pretrained VGG model is like Zeus's lightning bolt; it's so powerful that you can tackle almost any problem with it! 

Yes, but: PyTorch Hub's reliability has still not reached an impenetrable stage. Furthermore, it's still developing as a machine learning tool, so depending too much on it wouldn't be a great idea. 

Stay smart: Understand where you need to deep dive and train architectures from scratch and where you should be taking Torch Hub's help to speed things up. This is where understanding your problem statement, the data distribution, and other factors come into play. 

Click here to read the full tutorial

PyImageSearch University

This lesson is part of PyImageSearch University, our flagship program to help you master computer vision, deep learning, and OpenCV.  PyImageSearch University is updated each week with new lessons.

Don't know Python?  No problem, we've got you covered with a short and sweet Python course to get you going.

Having problems with your local development environment or IDE?  Fortunately, our pre-configured Colab Notebooks mean you can run code the moment you join PyImageSearch University. But, of course, you don't want to be a sys-admin, so don't waste time messing with your development environment.  

You can find the current lesson under Torch Hub 101 —  Practical Applications of Torch Hub,  and the direct link can be found here.

Want to Master Computer Vision and Deep Learning?

Do you think mastering computer vision and deep learning has to be time-consuming, overwhelming, and complicated? Or has to involve complex mathematics and equations? Or requires a degree in computer science?

That's not the case. All you need to master computer vision and deep learning is for someone to explain things to you in simple, intuitive terms. And that's exactly what we do. Our mission is to change education and how complex Artificial Intelligence topics are taught.

Inside PyImageSearch University, you'll find:

  • 30 courses on the hottest computer vision, deep learning, and OpenCV topics
  • 30 Certificates of Completion (one for each course)
  • 39+ hours of on-demand video
  • Pre-configured Jupyter Notebooks running in Google Colab
  • Run all code examples in your web browser — works on Windows, macOS, and Linux (no dev environment configuration required!)
  • Access to centralized code repos for all 500+ tutorials on the PyImageSearch blog
  • Easy one-click downloads for code, datasets, pre-trained models, etc.
  • Access on mobile, laptop, desktop, etc.
  • New courses released regularly and new tutorials weekly, ensuring you can keep up with state-of-the-art techniques

Click here to join PyImageSearch University


Adrian Rosebrock
Chief PyImageSearcher