Figure 1.

Illustration of StackBox framework. The framework builds distinct base learners (using the training data), and from these models predicts the bounding boxes around the detected objects. Using these predictions as input and using the ground-truth as output, a meta-model combines the base learners’ output, building a new model with improved performance. The base learners built previously are subsequently applied to the test data set to detect polyps on unseen data. Finally, these detections are used as input features for the meta-learner built on the training data to obtain the final predictions on the test data set.