EVENTO
Weakly Supervised Semantic Segmentation Applied to Remote Sensing: Seed Inference and Multi-class Balancing Strategies
Tipo de evento: Defesa de Dissertação de Mestrado
Obtaining comprehensive labeled data for remote sensing applications can be costly and time-consuming. So using traditional supervised learning can become infeasible [1]. This dissertation explores weakly supervised learning (WSL) as a compelling alternative, focusing on multi-class semantic segmentation with limited label availability [2]. More specifically, we delve into the realm of semantic segmentation for remote sensing imagery while using the ISPRS Potsdam dataset [3] as a benchmark for comparing our WSL approach with the fully supervised method. Furthermore, we address the significance of class imbalance challenges [4] and the necessity of inferring reliable and representative high confidence zones for the proper functioning of the proposed system, which are crucial for robust performance. To further improvement, we investigate a new loss term designed specifically for weakly supervised learning. This novel function effectively takes advantage of information previously discarded from unlabeled pixels.Para assistir acesse:meet.google.com/fgj-oqcb-oaj
Data Início: 17/04/2024 Hora: 14:30 Data Fim: 17/04/2024 Hora: 17:00
Local: LNCC - Laboratório Nacional de Computação Ciêntifica - Virtual
Aluno: Pascoassis Souza Santos Meira - - LNCC
Orientador: Gilson Antônio Giraldi - Laboratório Nacional de Computação Científica - LNCC
Participante Banca Examinadora: José Marcato Junior - Universidade Federal de Mato Grosso do Sul - UFMS Roberto Pinto Souto - Laboratório Nacional de Computação Científica - LNCC
Suplente Banca Examinadora: Bruno Richard Schulze - Laboratório Nacional de Computação Científica - LNCC Raul Queiroz Feitosa - Pontifícia Universidade Católica do Rio de Janeiro - PUC-RIO
