ImOV3D: Learning Open-Vocabulary Point Clouds 3D Object Detection from Only 2D Images

NeurIPS 2024 | Paper | Project Page | Code

Timing Yang*, Yuanliang Ju*, Li Yi
Shanghai Qi Zhi Institute, IIIS Tsinghua University, Shanghai AI Lab

Overview

ImOV3D is the first open-vocabulary 3D object detector trained entirely from 2D images — no 3D ground truth required. It bridges the 2D-3D modality gap via flexible modality conversion: lifting 2D images to pseudo point clouds (monocular depth estimation) and rendering point clouds back to pseudo images (ControlNet). This creates a unified image-PC representation for training a multimodal 3D detector.

Citation

@article{yang2024imov3d,
  title={ImOV3D: Learning Open Vocabulary Point Clouds 3D Object Detection from Only 2D Images},
  author={Yang, Timing and Ju, Yuanliang and Yi, Li},
  journal={Advances in Neural Information Processing Systems},
  volume={37},
  pages={141261--141291},
  year={2024}
}

Contact

Timing Yang: timingya@usc.edu · Yuanliang Ju: yuanliang.ju@mail.utoronto.ca