Optical Flow

 

The goal of this work was to compute optical flow information between video frames for frame interpolation.  This was achieved using an over-segmentation approach that simultaneously computed the flow vectors and image segmentations.  To increase the quality of the interpolation results matting information was also computed.  A detailed description of the algorithm can be found in our ICCV 2005 paper.

 


Here are several examples of frame interpolation using our algorithm:

 

forestSeq.avi

2 input frames with 15 frames interpolated.

Input images (zip)

 

fairSeg.avi

5 input frames with 5 frames interpolated between each pair.

 

 

 

    

woodsAll.wmv

2 input frames with 15 frames interpolated.

Input images (zip)

 

 

 

 

gardenSeq.avi

5 input frames with 5 frames interpolated between each pair.

 

 


Motion Blur

 

Here's an example video demonstrating exaggerated

motion blur using the above technique for frame interpolation.

 

    

blurfair.wmv