Abstract:
DPODv2, a dense correspondence-based three-stage 6 DoF object detector, is proposed. A 2D object detector, dense correspondence estimation network, and multi-view pose refinement method estimate a full 6 DoF pose. We propose a unified deep learning network that can use RGB or depth images, unlike monocular RGB-only methods. Differentiable rendering-based pose refinement is also proposed. Comparing predicted and rendered correspondences in multiple views yields a pose that matches all predicted correspondences. Our method is rigorously tested on different data modalities and training data in a controlled environment. RGB excels in correspondence estimation, while depth improves pose accuracy with good 3D-3D correspondences. Their partnership naturally performs best. We evaluate and ablate several difficult datasets to analyze and validate the results. DPODv2 performs well on all of them while remaining fast and scalable regardless of data modality or training data.
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
