point-transformer-v3-nuscenes-torch#

Point Transformer V3 backbone pretrained on nuScenes LiDAR semantic segmentation; produces embeddings for 3D point cloud samples.

Details

  • Model name: point-transformer-v3-nuscenes-torch

  • Model source: Pointcept/PointTransformerV3

  • Model author: Xiaoyang Wu, et al.

  • Model license: MIT

  • Model size: 528.83 MB

  • Exposes embeddings? yes

  • Tags: embeddings, 3d, point-cloud, lidar, nuscenes, torch, transformer, official

Requirements

  • Packages: torch, spconv-cu120, torch-scatter, timm, addict, open3d

  • CPU support

    • no

  • GPU support

    • yes

Example usage

 1import fiftyone as fo
 2import fiftyone.zoo as foz
 3
 4dataset = foz.load_zoo_dataset(
 5    "coco-2017",
 6    split="validation",
 7    dataset_name=fo.get_default_dataset_name(),
 8    max_samples=50,
 9    shuffle=True,
10)
11
12model = foz.load_zoo_model("point-transformer-v3-nuscenes-torch")
13
14dataset.apply_model(model, label_field="predictions")
15
16session = fo.launch_app(dataset)