Computer Vision for Data Scientists

Course details

Computer vision has come a long way since its humble beginnings. And today, it’s one of the most talked-about fields in tech. Join instructor Harpreet Sahota in this comprehensive overview of the history and evolution of this increasingly important industry, developing your understanding of convolutional neural networks, network training, deep learning models for image classification tasks, transfer learning with pretrained models, and more. Explore the wide variety of functionalities offered by the SuperGradients flexible training library, which gives you the power to shorten and streamline the model development lifecycle. Along the way, Harpreet shares practical insights on how to train models and networks more effectively, applying state-of-the-art techniques like exponential moving average, weighted average, batch accumulation, and BatchNorm.

Note: This course requires a basic working knowledge of machine learning as well as experience with Python and PyTorch.

Concepts

0. Introduction

01 - Computer vision introduction
02 - What you should know

1. Computer Vision

03 - What is computer vision
04 - A history of computer vision
05 - Limitations of traditional CV techniques
06 - ImageNet
07 - The deep learning revolution

2. Introduction to Convolutional Neural Networks

08 - Overview of CNNs
09 - Why CNNs
10 - Convolutional layers
11 - Types of convolutions
12 - Pooling layers
13 - Activation functions
14 - Fully connected layers

3. How Networks Are Trained

15 - Supervised learning and loss functions
16 - Backpropagation in CNNs
17 - Optimization techniques
18 - Regularization and data augmentation

4. The Evolution of CNN Architectures

19 - LeNet
20 - AlexNet
21 - VGG
22 - ResNet
23 - MobileNetV1
24 - MobileNetV2
25 - MobileNetV3
26 - EfficientNet

5. Transfer Learning

27 - Introduction to transfer learning
28 - Types of transfer learning
29 - Steps in feature extracting and fine-tuning
30 - Best practices for transfer learning

6. PyTorch Crash Course

31 - Setting up the environment
32 - Dataset and DataLoader
33 - Training setup
34 - The training loop
35 - Testing and evaluation
36 - Inference

7. Hands-on Transfer Learning with SuperGradients

37 - Introduction to SuperGradients
38 - The trainer
39 - Required training params
40 - Optional training params
41 - Training the model
42 - Predicting with the model
43 - How to solve almost any image classification problem with SG

8. Training Tricks

44 - Exponential moving average
45 - Weight averaging
46 - Batch accumulation
47 - Precise BatchNorm
48 - Zero weight decay on BatchNorm and bias
49 - Training tricks in SuperGradients

Conclusion

50 - Next steps