Context aware labeling of fashion items using graphical models
In the last years, artificial intelligence, and computer vision in particular, object detecion in images has experienced a revolution with the success of convolutional neural networks (CNNs) and the availability of massively parallel computing hardware at affordable prices.
With these convolutional neural networks CNNs, it is possible to have a very good detection rate for objects in an image. However, each object is detected individually, without taking into account the context of the whole picture, and the implications of other nearby detections.
In this project the student will define a graphical model to infer a context aware labeling of fashion items, using the results of a state-of-the-art fashion item detector. Manually labeled data will be used to learn the model that will combine the evidence provided by the detector to determine a final
coherent set of detections. This project will be done together with the company Wide Eyes Technologies. As a technology transfer project, there is the possibility of requesting a grant for outstanding students.
The requirements for this project are fluency in at least one of the programming languages commonly used in the computer vision community (primarily C/C++ or Python), and a strong mathematical background. Knowledge of graphical models and machine learning will be a plus. For any questions, please contact Arnau Ramisa, Jan Funke or Francesc Moreno-Noguer.
The work is under the scope of the following projects:
- RobInstruct: Instructing robots using natural communication skills (web)