Skip to content

SharperShape/eclair-dataset

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

16 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Research @ Sharper Shape (CVPRW 2024)

ECLAIR: A High-Fidelity Aerial LiDAR Dataset for Semantic Segmentation

This is the official repository of the ECLAIR dataset. For technical details, please refer to:

ECLAIR: A High-Fidelity Aerial LiDAR Dataset for Semantic Segmentation
Iaroslav Melekhov, Anand Umashankar, Hyeong-Jin Kim, Vlad Serkov, Dusty Argyle.
[Paper] [ECLAIR product page] [Download] [USM workshop@CVPR2024]

We introduce ECLAIR (Extended Classification of Lidar for AI Recognition), a new outdoor large-scale aerial LiDAR dataset designed specifically for advancing research in point cloud semantic segmentation. As the most extensive and diverse collection of its kind to date, the dataset covers a total area of 10km^2 with close to 600 million points and features eleven distinct object categories. To guarantee the dataset’s quality and utility, we have thoroughly curated the point labels through an internal team of experts, ensuring accuracy and consistency in semantic labeling

Semantic annotations (classes):

  • Ground: All points representing the Earth’s surface, including, soil, pavement, roads, and the bottom of water bodies.
  • Vegetation: All points representing organic plant life, ranging from trees, low shrubs, and tall grass of all heights.
  • Buildings: Man-made structures characterized by roofs and walls, encompassing houses, factories, and sheds
  • Transmission Wires: High-voltage wires for longdistance transmission from power plants to substations. Either directly connected to transmission towers or poles. Also includes transmission ground wires.
  • Distribution Wires: Lower-voltage overhead distribution wires distributing electricity from substation to end users. Includes span guy wires and communication wires.
  • Poles: Utility poles used to support different types of wires or electroliers. These can include poles with either transmission or distribution wires. Down guy wires, crossarms and transformers are also included in this class.
  • Transmission Towers: Large structures supporting transmission wires with the distinct characterisation of steel lattices and cross beams.
  • Fence: Barriers, railing, or other upright structure, typically of wood or wire, enclosing an area of ground.
  • Vehicle: All wheeled vehicles that can be driven.
  • Unassigned: A catch-all category for non-subject points. Anything that is not on the class list is classified as Unassigned. These include wooden pallets, trash, structures not large or strong enough to put under buildings (tents, boulders, etc.), and house antennas.

Running the demo

This demo will run a pre-trained model on the test split of the proposed dataset and compute the macro and per-class F1 score. You will be asked to provide a path to the model checkpoint (referred as $MODEL_WEIGHTS) and a dataset directory (referred as $DATA_DIR). Create them wherever you wish and make sure to provide their absolute paths.

  • Download the model weights (checkpoint) using this link and place it under $MODEL_WEIGHTS/model_eclair.pth
  • Download the ECLAIR dataset by filling in the following download form. The dataset is distributed under Creative Commons 4.0 (CC BY-NC-SA 4.0). Once downloaded and unzipped, one should get the following structure:
$DATA_DIR
|-- pointclouds
|       |-- pointcloud_1.laz
|       |-- pointcloud_2.laz
|       `-- ...
|-- labels.json
  • Run test.py as follows:
    python test.py --dataset_path $DATA_DIR --model_weights_file $MODEL_WEIGHTS/model_eclair.pth
    

Using the Docker image

  • Install Docker: First, make sure Docker is installed on your system. You can download it from Docker's official website and follow the installation instructions for your specific operating system.
  • Build the Docker Image: inside the repository, run:
    docker build -t eclair-image .
    
  • Run the Docker Container and mount the $DATA_DIR and $MODEL_WEIGHTS paths:
    docker run -it --rm -v $DATA_DIR:/data -v $MODEL_WEIGHTS:/model_weights eclair-image
    
  • Run the evaluation script inside the docker:
    python test.py --dataset_path /data --model_weights_file /model_weights/model_eclair.pth
    

Citation

If you find our work useful, please consider citing:

@inproceedings{eclair2024,
    title={ECLAIR: A High-Fidelity Aerial LiDAR Dataset for Semantic Segmentation},
    author={Melekhov, Iaroslav and Umashankar, Anand and Kim, Hyeong-Jin and Serkov, Vladislav and Argyle, Dusty},
    booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
    year={2024}
}

Copyright

The annotated dataset is licensed under Creative Commons 4.0 (CC BY-NC-SA 4.0). This data may not be used for commercial purposes. We will consider commercial usage on a case-by-case basis. Please contact us for more information.

Contact

For additional questions about the dataset or inquiries about commercial use, please contact [email protected]

About

Home of the open source ECLAIR dataset

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Contributors 4

  •  
  •  
  •  
  •