Skip to content

A labeled dataset from a subset of the MVSEC dataset for car detection at night driving conditions.

Notifications You must be signed in to change notification settings

SensorsINI/MVSEC-NIGHTL21

Repository files navigation

MVSEC-NIGHTL21

mvsec-nightl21-video_new.mp4

Citation

DOI

When use this dataset, please cite:

@InProceedings{Hu_2021_CVPR,
    author    = {Hu, Yuhuang and Liu, Shih-Chii and Delbruck, Tobi},
    title     = {v2e: From Video Frames to Realistic DVS Events},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
    month     = {June},
    year      = {2021},
    pages     = {1312-1321}
}

Parent dataset: MVSEC

MVSEC-NIGHTL21 is derived dataset of "The Multi Vehicle Stereo Event Camera Dataset" which is available here: https://daniilidis-group.github.io/mvsec/

Please also cite the original MVSEC paper:

  • Zhu, A. Z., Thakur, D., Ozaslan, T., Pfrommer, B., Kumar, V., & Daniilidis, K. (2018). The Multi Vehicle Stereo Event Camera Dataset: An Event Camera Dataset for 3D Perception. IEEE Robotics and Automation Letters, 3(3), 2032-2039.

Usage

Open In Colab

Data Description

MVSEC at night condition

We used outdoor_night1_data.hdf5 of the MVSEC dataset. The dataset is recorded with dual camera, we use the left camera. In the HDF5 archive, the relevant dataset can be accessed as following:

mvsec_data = h5py.File(mvsec_data_path, "r")

# raw frame
frame_data = mvsec_data["davis"]["left"]["image_raw"]
# frame timestamps
frame_ts = mvsec_data["davis"]["left"]["image_raw_ts"]

# raw events 
events_data = mvsec_data["davis"]["left"]["events"]

# event indices that corresponds to the frame
frame_event_inds = mvsec_data["davis"]["left"]["image_raw_event_inds"]

For visualization in this repository, we only used the raw frames.

MVSEC-NIGHTL21 Labels

In the validation set, there are 400 frames. The list of the frame indices is in frame_list.txt.

Among these 400 frames, 368 frames are labeled. The frames that don't have labels are listed in frames_that_dont_have_labels.txt. We labelled car in these frames.

The labelled groundtruths are stored in .txt files and can be found in mvsec_nightl21_labels.

Each labeled car is in the format car x_min y_min x_max y_max. For example:

car 48 112 143 170

means x_min=48, y_min=112, and x_max=143, y_max=170.

Visualization

  1. Install dependency

    pip install h5py
    pip install matplotlib
    pip install opencv-python
    
  2. Clone this repository

    git clone https://github.com/SensorsINI/MVSEC-NIGHTL21
    cd MVSEC-NIGHTL21
    
  3. Download the outdoor_night1_data.hdf5 from MVSEC dataset, available here

  4. Run The Visualization

     python visualize_mvsec_nightl21.py --mvsec_data /path/to/outdoor_night1_data.hdf5 --gt_root ./mvsec_nightl21_labels
    

    If everything works, you should see a video that annotates the cars.

Contact

Yuhuang Hu
yuhuang.hu@ini.uzh.ch

About

A labeled dataset from a subset of the MVSEC dataset for car detection at night driving conditions.

Topics

Resources

Stars

Watchers

Forks

Packages

No packages published

Languages