Qwen-3D: A Generalist 3D Vision-Language Model for Spatial Understanding

This repository contains the Qwen-3D model, as presented in Qwen-3D: A Generalist 3D Vision-Language Model for Spatial Understanding. Qwen-3D is a geometry-aware large multimodal model built on Qwen2.5-VL that introduces world-space tokens, 3D Rotary positional encoding, and a query-based mask decoder. It unifies referential grounding, instance segmentation, and visual question answering across images and videos.

Project page: https://qwen-3d.github.io/

Paper: https://arxiv.org/abs/2608.02980

Code: https://github.com/ll220/qwen3d

The checkpoint files for the 3B and 7B models qwen3d_3b.pth qwen3d_7b.pth are available in this repository. You can download it with huggingface-cli / hf, for example:

hf download katefgroup/Qwen-3D qwen3d_3b.pth
hf download katefgroup/Qwen-3D qwen3d_7b.pth

See the GitHub repository for code, loading instructions, and usage details.

Citation

@inproceedings{lin2026qwen3d,
  title     = {Qwen-3D: A Generalist 3D Vision-Language Model for Spatial Understanding},
  author    = {Lin, Lucy and Jain, Ayush and Liu, Yifan and Fragkiadaki, Katerina},
  booktitle = {European Conference on Computer Vision (ECCV)},
  year      = {2026}
}

License Note: This model is released under CC BY-NC 4.0. It may be used for non-commercial research purposes. Commercial use is not permitted under this license.