Abstract:
Convolutional neural networks trained end-to-end to regress dense disparities estimate stereo image depth with unmatched accuracy. If large amounts of labelled samples are available for training, this is possible for most tasks. Real-world applications rarely meet such an assumption, making adaptability crucial. We propose a deep stereo network adaptation paradigm for difficult and changing environments. We design a lightweight and modular architecture, M odularly AD aptive Net work (MADNet), and MAD and MAD++ algorithms to efficiently optimize independent sub-portions of the network. Our paradigm uses right-to-left image warping for self-supervision or stereo algorithms for online model adaptation. Both sources require only deployment-time input images. Thus, our network architecture and adaptation algorithms create the first real-time self-adaptive deep stereo system and enable a new paradigm for dense disparity regression end-to-end architectures.
Note: Please discuss with our team before submitting this abstract to the college. This Abstract or Synopsis varies based on student project requirements.
Did you like this final year project?
To download this project Code with thesis report and project training... Click Here
