blanchon/INRIA-Aerial-Image-Labeling

Inria Aerial Image Labeling Dataset

Description

The Inria Aerial Image Labeling Dataset is a building semantic segmentation dataset proposed in "Can semantic labeling methods generalize to any city? the inria aerial image labeling benchmark," Maggiori et al.. It consists of 360 high-resolution (0.3m) RGB images, each with a size of 5000x5000 pixels. These images are extracted from various international GIS services, such as the USGS National Map.

Project page: https://project.inria.fr/aerialimagelabeling/

Details

Structure

.
├── README.md
└── data
    ├── test
    │   └── images
    │       ├── bellingham1.tif
    │       ├── bellingham2.tif
    │       ├── ...
    │       └── tyrol-e36.tif
    └── train
        ├── gt
        │   ├── austin1.tif
        │   ├── austin2.tif
        │   ├── ...
        │   └── vienna36.tif
        └── images
            ├── austin1.tif
            ├── austin2.tif
            ├── ...
            └── vienna36.tif

Statistics

• Image Resolution: 0.3 meters per pixel
• Image Size: 5000x5000 pixels
• Total Images: 360
• Regions: 10 regions around the world, including both urban and rural areas.
• Split: Train and test sets are split into different cities for evaluating model generalization across dramatically different locations.
• Test Set Ground Truth Masks: Note that the ground truth masks for the test set have not been publicly released.

The dataset was originally used in the Inria Aerial Image Labeling Dataset Contest.

About the Dataset

The Inria Aerial Image Labeling Dataset is a comprehensive resource for semantic segmentation tasks in the field of remote sensing, with additional information as follows:

• Dataset Coverage: The dataset spans a total area of 810 km², meticulously divided into 405 km² for training and another 405 km² for testing purposes.

• Image Characteristics: This dataset offers aerial orthorectified color imagery, capturing scenes at an impressive spatial resolution of 0.3 meters per pixel.

• Semantic Classes: Ground truth data is provided for two fundamental semantic classes: "building" and "not building." It's important to note that ground truth data for the "not building" class is publicly disclosed exclusively for the training subset.

• Diverse Urban Settlements: The images cover a diverse range of urban settlements, ranging from densely populated areas such as San Francisco's financial district to picturesque alpine towns like Lienz in Austrian Tyrol.

• City-Based Split: Instead of merely dividing adjacent portions of the same images into the training and test subsets, this dataset adopts a unique approach. Different cities are included in each of the subsets. For instance, images from Chicago are part of the training set and excluded from the test set, while images from San Francisco are included in the test set and not in the training set. This design aims to assess the generalization capabilities of semantic labeling techniques across regions with varying illumination conditions, urban landscapes, and times of the year.

• Data Sources: The dataset was meticulously constructed by combining publicly available imagery and official building footprints.

This additional information further enriches the understanding of the Inria Aerial Image Labeling Dataset and its potential applications in remote sensing research.

Citation

If you use the Inria Aerial Image Labeling Dataset dataset in your research, please consider citing the following publication or the dataset's official website:

@article{xia2017aid,
  title     = {AID: A benchmark data set for performance evaluation of aerial scene classification},
  author    = {Xia, Gui-Song and Hu, Jingwen and Hu, Fan and Shi, Baoguang and Bai, Xiang and Zhong, Yanfei and Zhang, Liangpei and Lu, Xiaoqiang},
  journal   = {IEEE Transactions on Geoscience and Remote Sensing},
  volume    = {55},
  number    = {7},
  pages     = {3965-3981},
  year      = {2017},
  publisher = {IEEE}
}

AID: A Benchmark Dataset for Performance Evaluation of Aerial Scene Classification