You can find our dataset on predicting geoattributes using Mapillary street-level images in our github repository. code

We released our dataset on paired and unpaired cloudy and cloud-free satellite images for learning generative models to remove clouds in satellite images. For more information, please check our github repo. github

We released our dataset, Synthetic Aerial Vehicle Classification Dataset, to the research community. Our dataset consists of 55226 samples of 64x64 px images from an aerial platform generated by the Digital Imaging and Remote Sensing Software (DIRSIG). It is a remote sensing software developed by the Chester F. Carlson Center for Imaging Science at Rochester Institute of Technology. In the dataset, there are two classes : (1) vehicle, and (2) background. Overall, there are 27613 samples of vehicle images and 27613 samples of background images. The Ground Sampling Distance in this dataset are tuned to 0.3m on average as we target vehicle detection/classification in the Wide Area Motion Imagery (WAMI) platform. You can find some positive samples from the DIRSIG generated dataset and WAMI dataset in the figure below. Our goal by releasing this dataset is to avoid splitting a single video captured from the WAMI platform to form a training and validation dataset as it leads to overfitting to a certain scenario. Please cite our paper if you use this dataset for research purposes.

• DIRSIG generated training images and WAMI validation images link
  • The dataset is split into two folders :
    • train_dirsig that includes synthetic images and a text file containing the labels of the training samples.
    • validation_wami that includes 600 WAMI images and a text file containing the labels of the validation samples.

  @article{uzkent2017tracking,
      title={Tracking in Aerial Hyperspectral Videos using Deep Kernelized Correlation Filters},
      author={Uzkent, Burak and Rangnekar, Aneesh and Hoffman, Matthew J},
      journal={arXiv preprint arXiv:1711.07235},
      year={2017}
  }
  

We released our dataset, A Synthetic Aerial Hyperspectral Video for Vehicle Tracking, to the aerial tracking research community. This video contains 157 frames captured from an aerial platform generated by the Digital Imaging and Remote Sensing Software (DIRSIG). The average ground sampling distance in the video is 0.3m and the frame rate is set to be 1.42 fps. You can find the full video below. Our dataset also comes with the ground truth locations of the vehicles in the video. Please cite our paper if you this dataset for research purposes. For more information, we refer the readers to our paper shown below and our previous papers you can find in the publications section.

• Download Hyperspectral Frames link
• Download ReadMe and Ground Truth for Vehicles link
• Aerial Vehicle Tracking Video link

  @inproceedings{uzkent2016real,
    	title={Real-time vehicle tracking in aerial video using hyperspectral features},
    	author={Uzkent, Burak and Hoffman, Matthew J and Vodacek, Anthony},
   	booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops},
    	pages={36--44},
    	year={2016}
  }