1,449 pairs of aligned RGB and depth images captured from a variety of indoor scenes. The dataset includes dense per-pixel semantic labels and depth maps for every image, facilitating multi-modal computer vision research.
Use Cases
- Train monocular depth estimation models using the RGB images and depth map targets
- Develop semantic segmentation models using the RGB images and per-pixel class labels
- Implement multi-task learning architectures that simultaneously predict depth and semantic categories
Strengths
- 1,449 pairs of aligned RGB and depth images
- Dense per-pixel semantic labels for indoor objects and surfaces
- Data captured across 464 distinct indoor scenes using Microsoft Kinect