Scan2Cap: Context-aware Dense Captioning in RGB-D Scans

Dave Zhenyu Chen1      Ali Gholami2      Matthias Nießner1      Angel X. Chang2     

1Technical University of Munich       2Simon Fraser University


We introduce the task of dense captioning in 3D scans from commodity RGB-D sensors. As input, we assume a point cloud of a 3D scene; the expected output is the bounding boxes along with the descriptions for the underlying objects. To address the 3D object detection and description problems, we propose Scan2Cap, an end-to-end trained method, to detect objects in the input scene and describe them in natural language. We use an attention mechanism that generates descriptive tokens while referring to the related components in the local context. To reflect object relations (i.e. relative spatial relations) in the generated captions, we use a message passing graph module to facilitate learning object relation features. Our method can effectively localize and describe 3D objects in scenes from the ScanRefer dataset, outperforming 2D baseline methods by a significant margin (27.61% CiDEr@0.5IoUimprovement).



Qualitative results from baseline methods and Scan2Cap with inaccurate parts of the generated caption underscored. Our method (Scan2Cap) produces good bounding boxes with descriptions for the target appearance and their relational interactions withobjects nearby.


Paper | arXiv | Code

If you find our project useful, please consider citing us:

    title={Scan2Cap: Context-aware Dense Captioning in RGB-D Scans}, 
    author={Dave Zhenyu Chen and Ali Gholami and Matthias Nie├čner and Angel X. Chang},