vision.middlebury.edu
stereomviewMRFflowcolor
Optical Flow EvaluationDatasetsSubmit 

If you want to have your method evaluated, here's what to do:

  1. Go to the Datasets page and download one of the evaluation datasets. Use the color versions if your method can handle color, and use the multi-frame versions if your method can handle more than 2 frames.

  2. Adapt your flow algorithm so that it saves the resulting flow vectors using 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.

  3. Run your flow algorithm on each of the 12 sequences. Use constant parameters across all sequences. Use frame10.png as reference frame and compute the flow to frame11.png. Save the resulting flow as "flow10.flo" in each of the sequence directories. Also please keep track of the running time on the "Urban" dataset.

  4. Create a zip file with all of the .flo files from within the parent directory of all the sequence directories. Use your last name as the file name. For example: "zip smith.zip */flow10.flo".

  5. When you're ready to submit, go to the upload page. Be prepared to answer the following questions when submitting your results:
    1. Which version of the data did you use (# of frames, gray or color)?
    2. What kind of implementation (e.g., C/C++, Matlab)?
    3. What was the running time on the "Urban" data set, and what was the processor type and clock speed?
    4. Please also affirm that you used constant parameters for all datasets.
    Also be ready to provide bibliographic information about the paper describing your method. If you're submitting your paper anonymously to a conference, we will list your results as "anonymous".

  6. We will then evaluate your results and compile them in an online table. You will be able to see your results on a "private" page first, and we will only add your results to the public table after you give your consent.

 

 

 

 

 

 

 

 

 

 

Last modified: August 16 2009 by Daniel Scharstein