Video prediction network
Organization
Autonomous Driving Lab
Abstract
Training an end-to-end neural network to drive a car requires a lot of data, which needs to be collected using a specialized vehicle that can record the driving commands from a human driver. It is much easier to collect just general dashcam footage from cars. This footage could be used to train a network to predict the next frame in a self-supervised manner. Such a network presumably produces good representations for the downstream end-to-end driving task to make its driving decisions.
The project consists of:
1. Literature review of video prediction networks.
2. Collecting and preparing a dataset of driving videos.
3. Training a neural network to predict the next frame from a sequence of images.
4. Training an end-to-end driving neural network on top of the representations of the video prediction network.
The project consists of:
1. Literature review of video prediction networks.
2. Collecting and preparing a dataset of driving videos.
3. Training a neural network to predict the next frame from a sequence of images.
4. Training an end-to-end driving neural network on top of the representations of the video prediction network.
Graduation Theses defence year
2020-2021
Supervisor
Tambet Matiisen
Spoken language (s)
Estonian, English
Requirements for candidates
Level
Masters
Keywords
Application of contact
Name
Tambet Matiisen
Phone
5286457
E-mail