If your browser doesn't support automatic redirections please click here.

Supplementary Material Page for Our ICCV Paper

Modeling Temporal Coherence for Optical Flow

Automatic Redirection to the Pages of the Vision and Image Processing Group If your browser doesn't support automatic redirections please click here.


Additional Results

In the following tables, we show the reference frame, the four flow fields computed with our "joint spatial" approach and finally the trajectorial model map computed from these flow fields. Here, black marks no trajectorial regularization, gray marks second order regularization and white first order regularization. Note that for some sequences, only two frames are available (Dimetrodon, Venus, Teddy), so that only one flow field could be estimated. Trajectorial regularization becomes meaningless in this case.


Results on the Middlebury training data set










Left to Right: Reference frame, flow evolution, computed trajectorial model map.
Top to Bottom: Rhubberwhale, Hydrangea, Dimetrodon, Grove2, Grove3, Urban2, Urban3, Venus.



Results on the Middlebury training data set











Left to Right: Reference frame, flow evolution, computed trajectorial model map.
Top to Bottom: Army, Grove, Mequon, Schefflera, Urban, Wooden, Yosemite, Teddy



Visualisation of the motion trajectories over five frames


The following images show the computed trajectories for the six Middlebury training sequences. The background shows the reference frame. The corresponding pixel of each trajectory in this frame is highlighted with a big dot. Colors are chosen in accordance with the third flow field.



















Top to Bottom: Urban2, Urban3, Grove2, Grove3, Hydrangea, RubberWhale



Back to main page     Middlebury data set


Automatic Redirection to the Pages of the Vision and Image Processing Group If your browser doesn't support automatic redirections please click here.