Master Explainable AI: Interpreting Image Classifier Decisions/

Weakly Supervised Object Localization

Weakly Supervised Object Localization

Weakly supervised object localization (WSOL) focuses on localizing objects in an image with the supervision of image-level classification labels (obtained from a classification network). In this project, we will perform and evaluate WSOL using Class Activation Maps (CAMs) on the PASCAL-VOC dataset.

The project consists of two tasks, each comprising a code skeleton that we need to complete: 

  • Task 1: Localization: We will perform the WSOL using CAMs.

  • Task 2: Evaluation: We will evaluate the performance of CAM-based WSOL on the PASCAL-VOC dataset.

The following concepts/skills of Artificial Intelligence will be assessed in this project:

  • Class activation maps

  • Metrics of Interpretability

Estimated time of completion: 2 hours