Consecutive stereo image pairs with ground truth for scene labeling analysis (segmentation)
Sequence Length (Frames) |
2 |
Colour / Grey |
Grey |
Colour Depth |
12-bit |
Resolution (pixels) |
640 x 480 |
Stereo Rectified |
Yes |
Ego-motion data |
Yes |
Object data |
Yes |
- Consecutive stereo image pairs provided by Daimler AG (2010).
- Ground truth is only available for the second left camera in color and grey scale.
- Intrinsic and extrinsic camera parameters are available for download.
- The position of the tracked objects are available for download.
Sample Images & Downloads
- Each sequence is less than 1MB.
- Click here for a description of the ground truth data.
Intersection |
 |
INTERSECTION.zip (724 KB) |
An intersection scene with four oncoming cars. This scene contains partial occlusions, distant objects, as well as two nearby objects that move in the same direction with approximately equal velocity. |
Stroller |
 |
STROLLER.zip (830 KB) |
A pedestrian with a stroller is walking in front of a crossing car. The pose and motion state of the car is known from tracking. The pedestrian casts a strong stereo shadow on the object, i.e., there are large regions that can only be seen from one camera. |
Lead vehicle |
 |
LEAD_VEHICLE.zip (974 KB) |
The ego-vehicle follows the lead vehicle at approximately the same velocity through dense urban traffic, including two oncoming cars, a slow moving trailer ahead, and one car entering the visible field from the right. |
Reference:
Alexander Barth, Jan Siegemund, Annemarie Meissner, Uwe Franke, and Wolfgang Förstner, “Probabilistic Multi-Class Scene Flow Segmentation for Traffic Scenes“, in Proc. 32th Annual Symposium of the German Association for Pattern Recognition (DAGM ’10), 1-10, 2010. (click here to download this as a bib file).