TITAN - Honda Research Institute USA

TITAN Trajectory Inference using Targeted Action priors Network

Videos

Introduction

TITAN dataset is captured from a moving vehicle on highly interactive urban traffic scenes in Tokyo.

  • 700 video clips captured using a GoPro Hero 7 Camera, each with 10-20 seconds in duration
  • Images with a size of 1920 X 1200 pixels, annotated at 10 HZ sampling frequency
  • Synchronized odometry data obtained from an IMU sensor
  • 75,262 frames with 395,770 persons, 146,840 4-wheeled vehicles, and 102,774 2-wheeled vehicles
  • 50 labels including vehicle states and actions, pedestrian age groups, and targeted pedestrian action attributes

Data Distribution

 

 

Data Distribution

 

 

Action Hierarchy

 

 

Action Hierarchy

Data Format

DATA:

images_anonymized.tar
   |
   clip_xxx: sequence of clip
	|
	----images/yyyy.png (Camera DATA):

imu_data.tar
   |
   clip_xxx: sequence of clip
	|
	----synced_sensors_raw.csv (IMU DATA from GoPro): [image_ts, image_path, accel_ts, accel (m/s^2), gyro_ts, ang_vel (rad/s)]
	|
	----synced_sensors.csv (IMU DATA normalized bias from GoPro): [image_ts, image_path, accel_ts, accel (m/s^2), gyro_ts, ang_vel (rad/s)]

titan_0_4.tar (annotations)
   |
   clip_xxx.csv:(frame_id, label, obj_track_id, top, left, height, width, attr.Trunk, attr. Motion Status, attr.Doors Open, attr. Comunicative, attr.Complex, attr.Atomic, attr.Simple, attr.Transporting, attr.Age)


splits.tar
   |
   ----train_set.txt
   |
   ----val_set.txt
   |
   ----test_set.txt
   

Citation

This dataset corresponds to the paper, 'TITAN: Future Forecast using Action Priors', as it appears in the proceedings of Computer Vision and Pattern Recognition 2020. In the current release, the data is available for researchers from universities.

The process to obtain the dataset is as follows:

  • Please send an email at bdariush @ honda-ri dot com  using your official university email address.   
  • 'Data Sharing Agreement'  will be sent to you to be signed by a university representative (e.g. a university professor).
  • On review, instructions for downloading the dataset will be provided.

Please cite the following paper if you find this work useful:

@inproceedings{malla2020titan,
  title={TITAN: Future Forecast using Action Priors},
  author={Malla, Srikanth and Dariush, Behzad and Choi, Chiho},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={11186--11196},
  year={2020}
}