UniFusion: Unified Multi-view Fusion Transformer for Spatial-Temporal Representation in Bird’s-Eye-View
City-based split
This repo contains the "city-based" split proposed in the paper.
- Trianing set: the samples collected in
singapore-queenstownandsingapore-hollandvillage. - Validation set: the samples collected in
singapore-onenorthandboston-seaport.
The sizes of training and validation sets of "city-based" split are 26,093 and 8,056. For comparision, the sizes of the vanilla training and validation sets are 28,130 and 6,019.
We also provide a nuscenes_city_based_split.py to generate the city-based split.
Demo
We also provide a demo of our method.
