Non-central panorama indoor dataset

Bruno Berenguel-Baeta,Jesus Bermudez-Cameo,Jose J. Guerrero

DOI: https://doi.org/10.1016/j.dib.2022.108375

2024-01-30

Abstract:Omnidirectional images are one of the main sources of information for learning based scene understanding algorithms. However, annotated datasets of omnidirectional images cannot keep the pace of these learning based algorithms development. Among the different panoramas and in contrast to standard central ones, non-central panoramas provide geometrical information in the distortion of the image from which we can retrieve 3D information of the environment [2]. However, due to the lack of commercial non-central devices, up until now there was no dataset of these kinds of panoramas. In this data paper, we present the first dataset of non-central panoramas for indoor scene understanding. The dataset is composed by {\bf 2574} RGB non-central panoramas taken in around 650 different rooms. Each panorama has associated a depth map and annotations to obtain the layout of the room from the image as a structural edge map, list of corners in the image, the 3D corners of the room and the camera pose. The images are taken from photorealistic virtual environments and pixel-wise automatically annotated.

Databases,Computer Vision and Pattern Recognition

What problem does this paper attempt to address?

The main problem that this paper attempts to solve is the lack of annotated datasets applicable to non - central panoramas, especially datasets for indoor scene understanding. Specifically:

1. **Scarcity of datasets**: Existing learning - based scene understanding algorithms develop rapidly, but the corresponding annotated datasets (especially datasets of non - central panoramas) have not kept up with this development. Due to the lack of commercial non - central devices on the market, there were no publicly available datasets of this type before.

2. **Utilization of geometric information**: Different from traditional central panoramas, non - central panoramas provide 3D information of the environment through geometric distortion in the image. However, how to effectively utilize this geometric information for scene understanding remains a challenge.

To solve these problems, the author proposes the first indoor scene understanding dataset specifically for non - central panoramas. This dataset has the following features:

- **Rich annotation information**: Each non - central panorama is accompanied by a depth map and multiple annotation information, including structural edge maps, a list of corner points in the image, 3D corner points of the room, and camera pose.
- **Large - scale data volume**: The dataset consists of 2,574 RGB non - central panoramas, which are taken from approximately 650 different rooms.
- **Virtual environment generation**: All images are from realistic virtual environments and are obtained through automatic pixel - level annotation.

### Value of the dataset
- **Pioneering**: This is the first non - central panorama dataset, filling the gap in this field.
- **Algorithm development and evaluation**: Researchers can use this dataset to develop and evaluate new algorithms, especially in terms of geometric property utilization.
- **Adaptation of existing algorithms**: This dataset can also be used to adapt existing omnidirectional central image algorithms to non - central cases.

By providing such a comprehensive and high - quality dataset, the author hopes to promote the development of non - central panoramas in the fields of computer vision and indoor scene understanding.