CLIP2Point: Transfer CLIP to Point Cloud Classification with Image-Depth Pre-training.
Tianyu HuangBowen DongYunhan YangXiaoshui HuangRynson W. H. LauWanli OuyangWangmeng ZuoPublished in: CoRR (2022)
Keyphrases
- point cloud
- image classification
- image features
- input image
- supervised learning
- image data
- single image
- structure from motion
- training set
- similarity measure
- image segmentation
- training samples
- machine learning
- high resolution
- feature vectors
- feature extraction
- key frames
- laser scanner
- image matching
- hough transform
- nearest neighbor
- image pairs
- depth estimation
- normal vectors
- feature selection