Lab Instructor: Deep and Reinforcement Learning
Universitat Politècnica de Catalunya
Albert Mosella-Montoro is a PhD Candidate at Universitat Politècnica de Catalunya under the advisement of Professor Javier Ruiz-Hidalgo. He received a degree in Audiovisual Systems Engineering from Universitat Politècnica de Catalunya in 2015, after completing his thesis on Object Detection in Collision Path under the advisement of Professor Javier Ruiz-Hidalgo.
In 2017 he received a Master's degree in Computer Vision from Universitat Autònoma de Barcelona, after completing his thesis on Vehicle Detection using Instance Segmentation under the advisement of Professor Javier Ruiz-Hidalgo and Dr.-Ing Florian Baumman from Adasens Automotive GmbH.
Universitat Politècnica de Catalunya
Universitat Politècnica de Catalunya
Universitat Politècnica de Catalunya
Annual Catalan Meeting on Computer Vision
Universitat Politècnica de Catalunya
Universitat Politècnica de Catalunya
Annual Catalan Meeting on Computer Vision
PhD in Deep Learning
Universitat Politècnica de Catalunya
MSc in Computer Vision
Universitat Autonoma de Barcelona
BSc in Audiovisual Systems Engineering
Universitat Politècnica de Catalunya
Currently I am working on methodologies to process three-dimensional point clouds using deep learning techniques with the goal of understanding the information inherent on such data. To accomplish this, I am currently focusing on the following applications:
Furthermore, I am interested in prediction algorithms applied to different fields such as: weather pronostics, market studies, software testing and so on. Additionally, I am also interested in the speech recognition and emotion analysis fields.
Geometric 3D scene classification is a very challenging task. Current methodologies extract the geometric information using only a depth channel provided by an RGB-D sensor. These kinds of methodologies introduce possible errors due to missing local geometric context in the depth channel. This work proposes a novel Residual Attention Graph Convolutional Network that exploits the intrinsic geometric context inside a 3D space without using any kind of point features, allowing the use of organized or unorganized 3D data. Experiments are done in NYU Depth v1 and SUN-RGBD datasets to study the different configurations and to demonstrate the effectiveness of the proposed method. Experimental results show that the proposed method outperforms current state-of-the-art in geometric 3D scene classification tasks.
Convolutional neural networks (CNNs) have demonstrated their capability to solve different kind of problems in a very huge number of applications. However, CNNs are limited for their computational and storage requirements. These limitations make difficult to implement these kind of neural networks on embedded devices such as mobile phones, smart cameras or advanced driving assistance systems. In this paper, we present a novel layer named Hybrid Cosine Based Convolution that replaces standard convolutional layers using cosine basis to generate filter weights. The proposed layers provide several advantages: faster convergence in training, the receptive field can be increased at no cost and substantially reduce the number of parameters. We evaluate our proposed layers on three competitive classification tasks where our proposed layers can achieve similar (and in some cases better) performances than VGG and ResNet architectures.
I would be happy to talk to you if you need my assistance in your research or whether you need support for your company. I am always open to collaborating with the industry in order to do a knowledge transfer of my investigation and create products with it. If you are interested in a research collaboration, please use my academic e-mail to contact me. For other topics, please use the second e-mail provided. In both cases you can reach me in skype.