nickponline / dd-nerf-dataset

NerF computer vision dataset

DroneDeploy NeRF Dataset

DroneDeploy NeRF Dataset ("DNF") contains a number of real world scenes captured by drone that include high resolution images, camera poses (intrinsics and extrinsics) and dense color point clouds. It is intended for computer vision research. All reconstructions were created using DroneDeploy.

Information

Name	Image	Images	Points	Size	Preview
house1		220	3M	1.7GB	Pointcloud preview
house2		952	25M	8.6GB	Pointcloud preview
house3		182	5M	1.4GB	Pointcloud preview
house4		367	10M	5.2GB	Pointcloud preview
ruins1		391	6M	3.1GB	Pointcloud preview
ruins2		1174	56M	9.3GB	Pointcloud preview
ruins3		539	12M	4.4GB	Pointcloud preview
tower1		783	42M	6.1GB	Pointcloud preview
tower2		684	16M	5.4GB	Pointcloud preview
pipes1		98	7M	1.3GB	Pointcloud preview

Download

Download Links: https://drive.google.com/drive/folders/1kMHAS5l9Zy9yWHp6N4icm2xC7sIeThs7

Structure

house1/
   - cameras.xml
   - images
      - 00001.jpg
      - 00002.jpg
      - 00003.jpg
   - points
      - points.ply
house2/ ...
house3/ ...
  ...

The cameras.xml contains the poses of each camera along with the distortion parameters, intrinsics and extrinsics. See tools/util.py for working with this file directly. The images directory contains the images and points/points.ply contains the dense pointcloud with x,y,z,red,green,blue.

Tools

The tools directory contains an expanding collection of tools for working with and transforming the dataset. Contributions are welcome.

tools/dd2ngp.py - create a cameras.json ready for use with Instant Neural Graphics Primitives. The default cameras.json is included in each zip of the dataset and uses the 50 nearest images to the center of the scene. If you want to use different cameras you can use this script to re-generate a cameras.json.

Usage

instant-ngp$ ./build/testbed --scene ./datasets/house1/cameras.json

or

python scripts/run.py --mode="nerf" --scene="house1/cameras.json" --screenshot_transforms="house1/cameras.json"  --width="1216" --height="912" --screenshot_dir="house1/output" --near_distance="0"  --screenshot_spp="16"

License and Citation

@misc{Pilkington2022,
  author = {Nicholas Pilkington},
  title = {DroneDeploy NeRF Dataset},
  year = {2022},
  month = may
  publisher = {GitHub},
  journal = {GitHub repository},
  howpublished = {\url{https://github.com/nickponline/dd-nerf-dataset}},
}

This work is made available for academic use under CC BY-NC-SA. For commercial use and queries please contact [e-mail].