NITRDrone Dataset (NITRDD)

This is an aerial image dataset for semantic scene understanding and is under development.

LOG

Date	Log
2019.09.01	Initialization
2021.09.23	NITRDD(4) released

Dataset Overview:

-NITRDD

class	GT	(R, G, B)
background	0	(0, 0, 0)
road	1	(128, 0, 0)
occluded_road	2	(0, 128, 0)
vegetation	3	(128, 128, 0)

Directory Structure:

NITRDrone Dataset
│
└─── readme_images
|
└─────────data
│          |
│          └─── train
|          |       └───| images 
|          |       └───| gray_masks
|          |       └───| rgb_masks
|          └───   val
|          |       └───| images 
|          |       └───| gray_masks
|          |       └───| rgb_masks
|          |        
|          |    .gitignore
|          | 
|          |     frame271.jpg
|          ...
└───utils
│   │  label_generator.py
│   │  labels.txt 
│   │  requirements.txt
│   ...
│   
|   .gitignore
|    CITATION.cff
|    LICENSE
|    README.md
|   ...

DESCRIPTION:

Folder Name	Description
train	contains three folders as images, rgb_masks, gray_masks
val	contains three folders as images, rgb_masks, gray_masks
images	Aerial images captured by DJI Phantom 4 drone
rgb_masks	RGB masks of the corresponding images generated after annotations
gray_masks	Gray scale masks of the corresponding images
utils	Important files related to dataset preparation

Benchmark

The benchmark is released to look for better solutions for the proposed dataset (NITRDD). The performance metrics that are used OA (Overall Accuracy/pixelwise accuracy), mIoU (mean Intersection over Union). Researchers are welcomed to contribute new results!

-Obtained Results (NITRDD(4))

Models	#Parameters	OA	mIoU
FCN-8s	136M	0.72	0.16
FCN-16s	134M	0.83	0.68
FCN-32s	134M	0.86	0.63
FC_Densenet_103	9.42M	0.91	0.62
E-Net	350.65K	0.89	0.65
LinkNet	1.15M	0.93	0.46

Sample Images

image	mask

Download Link:

• NITRDrone(4): (train+val)
  GDrive

LICENSE:

This dataset is licenensed to Visual Surveillance Laboratory, Department of Computer Science and Engineering, National Institute of Technology, Rourkela, India.

Citation & Acknowledgements

This dataset is available publicly for only for non-commercial use. If you use this dataset in your research or wish to refer to the baseline results published in the README, please cite the work from Cite this repository as:

@article{nitrdd2021,
    title={The NITRDrone Dataset to Address the Challenges for Road Extraction from Aerial Images},
    author={Behera, Tanmay Kumar and Bakshi, Sambit and Sa, Pankaj Kumar and Nappi, Michael and Castiglione, Aniello and Vijayakumar, Pandi   and Gupta, Brij},        
    journal = {Journal of Signal Processing Systems},
    publisher={Springer},
    year={2021}, 
    note = "doi:10.1007/s11265-022-01777-0"
 }