Incremental Learning Techniques for Part-Based Semantic Segmentation

In this work we use an Incremental Learning approach to try to develop a model for the Part-Based Semantic Segmentation. Our framework uses two networks where the second one is fed with the output of the first. The first network performs the Semantic Segmentation task with 21 classes from the Pascal-VOC, while the second one performs the Part-Based Semantic Segmentation task with the Part-Based Semantic Segmentation with 108 classes/parts on the Pascal-Part dataset.