The dataset of ISIV (Incremental Segmentation for Indoor Video).
ISIV is an indoor Scene incremental segmentation dataset developed from the video segmentation dataset VSPW. The ISIV comprises 1,661 video segments (totaling 4,983 frames) of indoor scenes, encompassing 40 categories.
ISIV follows a two-tier structure, including:
- Structural Elements: Includes essential structural components of indoor scenes, such as walls, floors, and ceilings.
- Furnishings: Primarily consists of various functional furniture items, including tables, chairs, and sofas.
- Decorations: Comprises highly decorative items, such as lighting fixtures, paintings, and mirrors.
- Electronics: Specifically categorizes indoor electronic devices, including televisions and screens.
- Objects: Covers a range of small or everyday items used indoors, such as bottles, books, and food.
- Living Entities: Includes humans and plants, reflecting the biological elements within indoor environments.
A sample of the ISIV dataset is displayed below. There are three sets of data collected from a single video. Columns 1, 3, and 5 display the original images, while columns 2, 4, and 6 present the corresponding annotation information.
The download links of Google Drive and Baidu Netdisk:
| Link | |
|---|---|
| Google Drive | Download |
| Baidu Netdisk | Download |
Below is an overview of the directory and file structure:
indoor_dataset
├── img # The original images
│ ├── 1001_3iEIq5HBY1s_0.jpg # Example image file
│ ├── 1001_3iEIq5HBY1s_1.jpg
│ └─ ...
├── mask # The corresponding mask images
│ ├── 1001_3iEIq5HBY1s_0.jpg
│ ├── 1001_3iEIq5HBY1s_1.jpg
│ └─ ...
└─ proposal # The proposals
├── 1001_3iEIq5HBY1s_0.jpg
├── 1001_3iEIq5HBY1s_1.jpg
└─ ...
If this codebase is useful to you, please cite our work:
@misc{ISIV2024,
title = {Incremental Segmentation for Indoor Video (ISIV) Dataset},
author = {Zekang Zhang and Xiao Yu and Hongguang Zhu and Yan Fang and Yunchao Wei and Yao Zhao},
institution = {The Institute of Information Science, Beijing Jiaotong University},
howpublished = {\url{https://github.com/hhhhh12131/ISIV}},
year = {2024}
}