Publications

1. Gomes, P., Rossi, S., & Toni, L. (2022). Explaining Hierarchical Features in Dynamic Point Cloud Processing. ArXiv Preprint ArXiv:2209.15557.
   PDF BIB ABSTRACT

   @article{gomes2022explaining,
     title = {Explaining Hierarchical Features in Dynamic Point Cloud Processing},
     author = {Gomes, Pedro and Rossi, Silvia and Toni, Laura},
     journal = {arXiv preprint arXiv:2209.15557},
     year = {2022},
     file = {2209.15557.pdf}
   }
   

   This paper aims at bringing some light and understanding to the field of deep learning for dynamic point cloud processing. Specifically, we focus on the hierarchical features learning aspect, with the ultimate goal of understanding which features are learned at the different stages of the process and what their meaning is. Last, we bring clarity on how hierarchical components of the network affect the learned features and their importance for a successful learning model. This study is conducted for point cloud prediction tasks, useful for predicting coding applications.

2. Gomes, P., Rossi, S., & Toni, L. (2021). Spatio-temporal Graph-RNN for Point Cloud Prediction. 2021 IEEE International Conference on Image Processing (ICIP), 3428–3432.
   PDF BIB ABSTRACT

   @inproceedings{gomes2021spatio,
     title = {Spatio-temporal Graph-RNN for Point Cloud Prediction},
     author = {Gomes, Pedro and Rossi, Silvia and Toni, Laura},
     booktitle = {2021 IEEE International Conference on Image Processing (ICIP)},
     pages = {3428--3432},
     year = {2021},
     file = {spatio-temporal-graph-rnn-for-point-cloud-prediction-3.pdf},
     organization = {IEEE}
   }
   

   In this paper, we propose an end-to-end learning network to predict future frames in a point cloud sequence. As main novelty, an initial layer learns topological information of point
   clouds as geometric features, to form representative spatiotemporal neighborhoods. This module is followed by multiple Graph-RNN cells. Each cell learns points dynamics (i.e.,
   RNN states) by processing each point jointly with the spatiotemporal neighbouring points. We tested the network performance with a MINST dataset of moving digits, a synthetic
   human bodies motions and JPEG dynamic bodies datasets. Simulation results demonstrate that our method outperforms baseline ones that neglect geometry features information

3. Gomes, P. (2021). Graph-based Network for Dynamic Point Cloud Prediction. Proceedings of the 12th ACM Multimedia Systems Conference, 393–397.
   BIB

   @inproceedings{gomes2021graph,
     title = {Graph-based Network for Dynamic Point Cloud Prediction},
     author = {Gomes, Pedro},
     booktitle = {Proceedings of the 12th ACM Multimedia Systems Conference},
     pages = {393--397},
     year = {2021}
   }
   

4. Gomes, P., & Silva Cruz Luı́s A. (2019). Pseudo-sequence light field image scalable encoding with improved random access. 2019 8th European Workshop on Visual Information Processing (EUVIP), 16–21.
   BIB ABSTRACT

   @inproceedings{gomes2019pseudo,
     title = {Pseudo-sequence light field image scalable encoding with improved random access},
     author = {Gomes, Pedro and da Silva Cruz, Lu{\'\i}s A},
     booktitle = {2019 8th European Workshop on Visual Information Processing (EUVIP)},
     pages = {16--21},
     year = {2019},
     organization = {IEEE}
   }
   

   A new form of digital images, light field images contain considerably more information than conventional 2D digital images. However, light field images require significantly more storage, transmission and processing resources than conventional 2D images, calling for efficient compression schemes. In this work, light field images are interpreted as multi-view sequences that are compressed using MV-HEVC. A novel scan order based on blocks of sub-aperture images and a two dimensional prediction scheme are proposed for improved random access performance, as well as to provide a scalable representation of the light field data. Our proposed compression scheme achieves PSNR gains as high as 7.8 dB over reference coding schemes based on JPEG.