EVAL_PAN1555-SPP/HDP Panasonic

Submission The EVAL_PAN1555-SPP/HDP part of Panasonic Submission is a software package that implements the evaluation protocol of the PAN1555 Challenge, using advanced deep learning architectures. The package consists of two components: the SPP (superpixel-based segmentation and classification) and the HDP (hierarchical deep learning-based segmentation and classification). The SPP component applies to the segmentation task an approach based on Superpixels, taking into consideration both global context and local characteristics. The HDP component implements a novel hierarchical deep learning-based approach for segmentation tasks. This approach combines an unsupervised phase and a supervised learning phase, which produces better results compared to approaches that rely only on supervised learning or on unsupervised learning approaches. With the EVAL_PAN1555-SPP/HDP software package, users are able to evaluate the performance of a given approach to the segmentation task provided by the PAN1555 Challenge.