TOD3Cap: Towards 3D Dense Captioning in Outdoor Scenes

Bu Jin1,2, Yupeng Zheng1,2*, Pengfei Li3, Weize Li3, Yuhang Zheng4, Sujie Hu3,
Xinyu Liu5, Jinwei Zhu3, Zhijie Yan3, Haiyang Sun2, Kun Zhan2, Peng Jia2,
Xiaoxiao Long6, Yilun Chen3, Hao Zhao3,
(*Indicates Corresponding Author)
Contact: jinbu18@mails.ucas.ac.cn, zhengyupeng2022@ia.ac.cn

1 CASIA, 2 Li Auto, 3 AIR, Tsinghua University, 4 Beihang University, 5 HKUST, 6 HKU,

Paper Data Code

Abstract

3D dense captioning stands as a cornerstone in achieving a comprehensive understanding of 3D scenes through natural language. It has recently witnessed remarkable achievements, particularly in indoor settings. However, the exploration of 3D dense captioning in outdoor scenes is hindered by two major challenges: 1) the domain gap between indoor and outdoor scenes, such as dynamics and sparse visual inputs, makes it difficult to directly adapt existing indoor methods; 2) the lack of data with comprehensive box-caption pair annotations specifically tailored for outdoor scenes. To this end, we introduce the new task of outdoor 3D dense captioning. As input, we assume a LiDAR point cloud and a set of RGB images captured by the panoramic camera rig. The expected output is a set of object boxes with captions. To tackle this task, we propose the TOD3Cap network, which leverages the BEV representation to generate object box proposals and integrates Relation Q-Former with LLaMA-Adapter to generate rich captions for these objects. We also introduce the TOD3Cap dataset, the largest one to our knowledge for 3D dense captioning in outdoor scenes, which contains 2.3M descriptions of 64.3K outdoor objects from 850 scenes in nuScenes. Notably, our TOD3Cap network can effectively localize and caption 3D objects in outdoor scenes, which outperforms baseline methods by a significant margin (+9.6 CiDEr@0.5IoU).

Contributions

  • We introduce the outdoor 3D dense captioning task to densely detect and describe 3D objects, using LiDAR point clouds along with a set of panoramic RGB images as inputs. Its unique challenges are highlighted in Fig.1.
  • We provide the TOD3Cap dataset containing 2.3M descriptions of 63.4k instances in outdoor scenes and adapt existing state-of-the-art approaches on our proposed TOD3Cap dataset for benchmarking.
  • We show that our method outperforms the baselines adapted from representative indoor methods by a significant margin (+9.6 CiDEr@0.5IoU).

Teaser of TOD3Cap

main_graph

Fig1: We introduce the task of 3D dense captioning in outdoor scenes (right). Given point clouds (right middle) and multi-view RGB inputs (right top), we predict box-caption pairs of all objects in a 3D outdoor scene. There are several fundamental domain gaps (middle column) between indoor and outdoor scenes, including Status, Point Cloud, Perspective, and Scene Area, bringing new challenges specific to outdoor scenes. Meanwhile, our outdoor 3D dense captioning (right bottom) contains more comprehensive concepts than indoor scenes (left bottom).

Pipeline of TOD3Cap

main_graph

Fig2: Architecture of our proposed TOD3Cap network. Firstly, BEV features are extracted from 3D LiDAR point cloud and 2D multi-view images, followed by a query-based detection head that generates a set of 3D object proposals from the BEV features. Secondly, to capture the relationship information, we utilize a Relation Q-Former where the objects interact with other objects and the surrounding environment to get the context-aware features. Finally, with an Adapter, the features are processed to be prompts for the language model to generate dense captions. This formulation does not require a re-training process of the language model.

Qualitative Results

main_graph

Fig3: Qualitative results for our proposed TOD3Cap network. In the top left, we show our predicted bounding boxes and corresponding captions in the first row and ground truth in the second row. In the top right, we show our predicted bounding boxes in blue and the ground truth bounding boxes in red. In the bottom, we mark the wrong descriptions in red. The TOD3Cap network produces impressive results except for a few mistakes.

Acknowledgement

We would like to thank Dave Zhenyu Chen at Technical University of Munich for his valuable proofreading and insightful suggestions. We would also like to thank Lijun Zhou and the student volunteers at Li Auto for their efforts in building the TOD3Cap dataset.

BibTeX

@article{jin2024tod3cap,
      title={TOD3Cap: Towards 3D Dense Captioning in Outdoor Scenes},
      author={Jin, Bu and Zheng, Yupeng and Li, Pengfei and Li, Weize and Zheng, Yuhang and Hu, Sujie and Liu, Xinyu and Zhu, Jinwei and Yan, Zhijie and Sun, Haiyang and others},
      journal={arXiv preprint arXiv:2403.19589},
      year={2024}}