Ofertes treballs fi d'estudishttps://eseiaat.upc.edu/ca/curs-actual/treballs-fi-estudis/ofertes-treballs-fi-destudishttps://eseiaat.upc.edu/++resource++plone-logo.svg
The deployment of arotificial intelligence is currently hampered by the complexity of generating large enough datasets in real conditions which, in addition, are valid to cover the different scenarios required for the perception problem being solved. A potential solution gaining acceptance in the community is the use of synthetic data generated from simulation environments. However, a significant challenge remains in creating realistic simulated data that match the specifications of the sensor and accurately recreate the behaviour of light in the environment. Within this project we will enhance the capabilities of synthetic dataset generation in well stablished applications (such as NVIDIA Omniverse or CARLA) adding to them a layer of physics which turns them into more robust, realistic simulations of reality.
Objectius:
To explore synthetic dataset generation and to improve its current performance adding physics layers on it (such as radiometry, colour, diffusivity, etc).
To generate synthetic datasets in commercial environments.
To test the performance of traning using such datasets (physics-enabled and not physics-enabled) to validate the usefulness of the approach and quantify the improvement brought on in computer vision or perception tasks.
Tasques a realitzar / Característiques:
Get involved in synthetic dataset generation tools, to evluate how to fit in and out physics parameter sin the environments.
Introduce physics-based algorithms for density, sampling, and spatial resolution which add on to existing algorithms
To generate synthetic datasets based on such algorithms and to compare them with existing ones in traning an AI for perception tasks.
Tema:
ENGINYERIA AUDIOVISUAL
Tipus:
Projecte
Càrrega de treball:
24 ECTS
Requisits:
Intermediate programming skils in MAtlab/Python/C++
Fluent English
Independent, self motivated student.
Comparteix: