Skip to content

Week 4: Transfer learning and object detection

This week introduces students to training models, as well as object detection pipelines and metrics. Students will learn about transfer learning with an image classification problem (learning the difference in a hands-on way, between feature extraction and fine-tuning). The week concludes with demo on object detection and illustrate metrics on object detection pipelines (mAP, IoU).

Focus

  • Understand what transfer learning is and why it’s useful
  • Fine-tune a pretrained model for a new classification task
  • Explore object detection with a pretrained model

Hands-On Activities

  • Load a pretrained model (e.g., ResNet18 or MobileNetV2) and fine-tune it (e.g., on fashion mnist).
  • Evaluate and visualize model performance using metrics from week 3.
  • Run an object detection model on custom images using pretrained models

Learning Outcomes

By the end of this week, students will be able to:

  • Explain what transfer learning is and when to use it
  • Modify and fine-tune pretrained image classification models (unfreezing different levels of the model: fully connected layer vs some of the feature extraction layers).
  • Evaluate performance while tracking overfitting on small datasets
  • Compare classical ML and deep learning workflows
  • Run an object detection model using a pretrained network
  • Explain metrics on object detection networks

Resources

Instructor Notes

Students should focus on understanding the logic of training and transfer learning rather than all the details: we are training future data engineers not getting in the weeds with the ML bits.