Titulo Estágio
Stereoscopic Vision with Deep Learning
Áreas de especialidade
Sistemas Inteligentes
Local do Estágio
DEI-FCTUC
Enquadramento
Computer vision has had enormous progress in the last couple of years mainly due to usage of deep learning techniques such as deep convolutional networks — both Google and Facebook have been resorting to this kind of technique for their facial recognition software and other image processing tools.
Image processing can be useful for many practical applications such as scene classification, object recognition, and decision making.
However, comparatively, stereoscopic vision (depth perception from pairs of 2d images) has had less attention from the research community.
Perception of the third dimension (depth) from 2d images can be used to improve the quality and efficiency of tasks such as image segmentation, object recognition and background filtering; these techniques can be applied in modern autonomous vehicles (e.g., cars, drones, etc.) gifted with binocular image capturing devices.
Objetivo
The goal of this project is to develop a computational model for stereoscopic vision that is capable of calculating the depth (distance from observer) of the objects in a scene, from pairs of 2d images.
The model will be built resorting to a deep architecture of neural network models, such as Convolutional Neural Networks or Stacked-Autoencoders; it will be trained with data sets such as the ones freely available at the Middlebury College’s website.
Plano de Trabalhos - Semestre 1
- Problem analysis, identification of usable libraries and frameworks for the development of the solution, and characterisation of the technical limitations of the hardware available at DEI [September-October 2016]
- State-of-the-art analysis in Computer Stereoscopic Vision [October 2016]
- Definition of detailed objectives, requirements, and architecture [October-November 2016]
- Implementation of the first prototype [November-December 2016]
- Training and testing of the first prototype [December 2016]
- Writing of the intermediate report [October-December 2016]
Plano de Trabalhos - Semestre 2
- Reflexion on the intermediate evaluation; integration of the comments by the jury [January 2017]
- Definition and implementation of the second version of the model [February-March 2017]
- Training and testing of the second version of the model [March-April 2017]
- Writing and submission of a scientific paper [March-April 2017]
- Writing of the final report [Fevereiro - Junho 2016]
Condições
This is a non-paid internship that will take place at DEI-FCT-UC, and the student will use his/her own laptop.
Applicant students should comply with the following minimum requirements:
Soft Skills:
- Strong interest in artificial intelligence/machine learning/computer vision
- Pro-active
- Strong english knowledge
- Dynamic
Hard Skills:
- Very good Python / Matlab programming skills
- Solid mathematical knowledge
Observações
Orientadores: Alexandre Miguel Pinto, Bernardete Ribeiro
Email Orientador(es): ampinto@dei.uc.pt, bribeiro@dei.uc.pt
Telefone Orientador: 239 790 069, 239 790 087
Orientador
Alexandre Miguel Pinto
ampinto@dei.uc.pt 📩