Activity Recognition

Sports Activity Recognition using Transfer Learning

Objective: Build a knn-classifier to recognize sports activities using Transfer Learning..

Why Transfer Learning? While pre-trained models are great for recognizing cats & dogs, or cars & trucks, here, I show you how to customize a pre-trained model for a new task in 3 simple steps.

  1. Compile a training video to demonstrate various activities you want to capture automatically using Transfer Learning
  2. For each activity, capture few screenshots and tag them manually to train the classifier
  3. Run predictions to recognize activities in the new test-video

Because the pre-trained model (used for feature extraction) is already trained on a very large dataset, the classifier achieves decent accuracy even though the new training sample used for customization is relatively much smaller in size.