This is the JAAD dataset aiming to provide series of samples for pedestrian and car detection,
action and gesture recognition and behavioral study of traffic participants. The data consists
of 346 high-resolution video clips (5-15s) with annotations showing various situations typical
for urban driving. The videos are recorded using dashboard mounted cameras.

All video sequences are available in mp4 and seq format (requires Piotr Dollar's toolbox).

CAMERA INFORMATION:
Note that three cameras were used for data collection.
videos 1-60: GoPro Hero+ (1920x1980)
videos 61-70: Highscreen Box Connect (1280x720)
videos 71-346: Garmin GDR-35 (1920x1980)
NOTE: videos #28, #37, #38, #39, #40, #41, #52, #53 have a higher frame rate (fps=60) than the rest of the dataset (fps=30).

Annotations for the dataset can be downloaded from our GitHub repo (https://github.com/ykotseruba/JAAD_dataset). There you can also find all relevant information regarding the types and formats of the available annotations.

Please send email to Yulia Kotseruba (yulia_k@cse.yorku.ca) or Amir Rasouli (aras@cse.yorku.ca) if there are any problems with downloading or using the data.