01 · Prepare the images
02 · Turn predictions into labels
03 · Learning and regularization
04 · Classify the new scene
Labelled renderingSource · known category: bicycle
Street photographTarget · category not supplied
Prediction → pseudo-labelExample: motorcycle
Source · joint learningPseudo-label generation ↔ optimization
Target · regularized learningInitial training
Category predictionUse the adapted classifier
Source images + known labels
Updated model → later predictions