TV-L1 Optical Flow Estimation
Please cite the reference article if you publish results obtained with this online demo.

Select Data

Click on an image pair to use it as the algorithm input

With ground truth
Homography
Yosemite no clouds
Rotation
Yosemite clouds
Half pixel
One pixel
Identity
Ten pixels
Marbeblocks
Book
Sinxy
Stmichel
Affinity
Vcbox
Blocks
Sphere
Complex
Street
Dimetrodon
Grove2
Grove3
Hydrangea
Rubber whale
Urban2
Urban3
Venus
Without ground truth
Army
Backyard
Basketball
Dumptruck
Evergreen
Grove
Mequon
Schefflera
Teddy
Urban
Wooden
Ettlinger
Rheinhafen
Latour, image
Latour, gradient
Fortuny, image
Fortuny, gradient
Pluto
Adriatic

Upload Data

Upload a pair of images to use as the algorithm input (and, optionally, a ground truth for comparison).




The ground truth must specify a two-dimensional vector field, for example a .flo file (as in the Middlebury database) or a TIFF file with two floating-point channels.