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

TL;DR

This paper presents Scan2Cap, a novel end-to-end method for dense captioning of 3D scenes from RGB-D scans, integrating object detection, description, and spatial relation modeling to improve accuracy.

Contribution

The paper introduces Scan2Cap, a new approach that combines attention and graph modules for 3D object detection and natural language description in RGB-D scans.

Findings

Outperforms 2D baseline methods by 27.61% in CiDEr@0.5IoU

Effectively localizes and describes objects in 3D scenes

Utilizes message passing for modeling object relations

Abstract

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).