CAL performs zero-shot, open-vocabulary panoptic scene completion from a single Lidar scan using a model distilled from pseudo-labels mined from temporal video and Lidar sequences.
This MLP block consists of layers with dimensions [384, 512, 1024, 768, 768], where the final dimension, 768, is the dimensionality of the CLIP embedding space
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Towards Learning to Complete Anything in Lidar
CAL performs zero-shot, open-vocabulary panoptic scene completion from a single Lidar scan using a model distilled from pseudo-labels mined from temporal video and Lidar sequences.