vision.middlebury.edu
stereomviewMRFflowcolor
Optical Flow EvaluationDatasetsSubmit 

Evaluation Datasets

With hidden ground-truth flow

Hidden Texture Synthetic Stereo High-speed camera (no GT)

Army

Mequon

Schefflera

Wooden

Grove

Urban

Yosemite

Teddy

Backyard

Basketball

Dumptruck

Evergreen
# frames 8 8 8 8 8 8 8 2 8 8 8 8
Flow Eval yes yes yes yes yes yes yes yes -- -- -- --
Interp Eval -- yes yes -- -- yes -- yes yes yes yes yes

Download:ColorGray
All frames (usually 8)   eval-color-allframes.zip (31MB) eval-gray-allframes.zip (12MB)
Two frames only eval-color-twoframes.zip (8MB) eval-gray-twoframes.zip (3MB)

Instructions for participating in the evaluation can be found on the Submit page.


Other Datasets

With public ground-truth flow. These can be used for training.

Hidden Texture Synthetic Stereo High-speed camera (no GT)

RubberWhale

Hydrangea

Dimetrodon

Grove2

Grove3

Urban2

Urban3

Venus

Beanbags

DogDance

MiniCooper

Walking
# frames 8 8 2 8 8 8 8 2 8 8 8 8

Download:ColorGray
All frames (usually 8)   other-color-allframes.zip (32MB) other-gray-allframes.zip (12MB)
Two frames only other-color-twoframes.zip (9MB) other-gray-twoframes.zip (3.4MB)
Ground-truth flow other-gt-flow.zip (13MB)
Ground-truth interpolation other-gt-interp.zip (4.5MB) other-gt-interp-gray.zip (1.7MB)

The ground-truth flow is provided in our .flo format. Information and C++ code is in flow-code.zip, which contains the file README.txt. A Matlab version is available in flow-code-matlab.zip.


Acknowledgments

Many thanks to Brad Hiebert-Treuer and Alan Lim, who spent countless hours creating the hidden texture datasets. Thanks also to 3delight, Mental Images, Pixar, Doug Creel, and Luca Fascione for their help with the synthetic data.

 

 

 

 

 

 

 

 

 

 

Last modified: August 17 2009 by Daniel Scharstein