README.md

metadata

license: mit
library_name: torch-pointcloud
tags:
  - point-cloud
  - 3d
  - pytorch
  - torch-pointcloud
  - act
  - classification
datasets:
  - scanobjectnn
base_model: torch-pointcloud/act.pretrain.runpei-dong
model-index:
  - name: act.scanobjectnn-objonly.runpei-dong
    results:
      - task:
          type: point-cloud-classification
        dataset:
          name: ScanObjectNN (OBJ_ONLY)
          type: scanobjectnn
        metrics:
          - name: OA
            type: accuracy
            value: 91.22
          - name: mAcc
            type: accuracy
            value: 90.26

Model card for act.scanobjectnn-objonly.runpei-dong

An ACT point cloud classification model (masked point transformer distilled from a 2D image transformer). Trained on ScanObjectNN (OBJ_ONLY).

Model Details

• Model Type: Point cloud classification
• Model Stats:
  • Params (M): 22.1
  • Classes: 15
  • Features: 768
• Dataset: ScanObjectNN (OBJ_ONLY)
• Metrics: OA 91.22, mAcc 90.26
• Paper: Autoencoders as Cross-Modal Teachers: Can Pretrained 2D Image Transformers Help 3D Representation Learning?
• Converted from: RunpeiDong/ACT (MIT)
• Library: torch-pointcloud

Install

pip install torch-pointcloud

Usage

import torch
import torch_pointcloud as tp
from torch_pointcloud.utils.data import collate

model, info = tp.create_model(
    "act.scanobjectnn-objonly.runpei-dong",
    pretrained=True,
    return_info=True,
)
model = model.eval()

# synthetic sample with the keys a dataset provides
num_points = 8192
sample = {
    "pos": torch.randn(num_points, 3),
}
data = collate([sample])

with torch.no_grad():
    logits = model(data.get("x"), data["pos"], data["batch"])

Feature extraction

with torch.no_grad():
    embeddings = model.forward_features(data.get("x"), data["pos"], data["batch"])

model.reset_classifier(num_classes=0)
with torch.no_grad():
    embeddings = model(data.get("x"), data["pos"], data["batch"])  # (B, 768)

Citation

@inproceedings{dong2023act,
  title     = {Autoencoders as Cross-Modal Teachers: Can Pretrained {2D} Image Transformers Help {3D} Representation Learning?},
  author    = {Dong, Runpei and Qi, Zekun and Zhang, Linfeng and Zhang, Junbo and Sun, Jianjian and Ge, Zheng and Yi, Li and Ma, Kaisheng},
  booktitle = {ICLR},
  year      = {2023}
}

@inproceedings{uy2019scanobjectnn,
  title   = {Revisiting Point Cloud Classification: A New Benchmark Dataset and Classification Model on Real-World Data},
  author  = {Mikaela Angelina Uy and Quang-Hieu Pham and Binh-Son Hua and Duc Thanh Nguyen and Sai-Kit Yeung},
  booktitle = {ICCV},
  year    = {2019}
}

@software{dujardin2026pytorchpointcloud,
  author  = {Arthur Dujardin},
  title   = {PyTorch PointCloud},
  year    = {2026},
  doi     = {10.5281/zenodo.22159632},
  url     = {https://github.com/arthurdjn/pytorch-pointcloud},
}