ScanEnts3D: Exploiting Phrase-to-3D-Object Correspondences for Improved Visio-Linguistic Models in 3D Scenes.
Ahmed AbdelreheemKyle OlszewskiHsin-Ying LeePeter WonkaPanos AchlioptasPublished in: CoRR (2022)
Keyphrases
- d objects
- d scene
- cad models
- viewpoint
- multiple views
- partial matching
- multi view
- pose estimation
- range data
- object recognition
- geometric models
- depth map
- planar surfaces
- shape descriptors
- line drawings
- camera parameters
- surface patches
- image based rendering
- three dimensional
- point correspondences
- three dimensional objects
- training data
- camera calibration
- single image
- active learning
- optical flow