Efficient Hierarchical Graph Based Video Segmentation - WebService

This is a web-service we offer to other researchers to try out our video segmentation algorithm. You can upload a video and after server-side processing you are able to download the spatio-temporal segmentation as well as some videos with different levels of the segmentation overlayed.

Use for academic research only, no large-scale or commerical use!

Visit our project site for details and example videos
News
  • 9/20/2010: Sample program updated for improved build support on linux and windows.
  • 7/1/2010: Sample program to read the segmentation is now available.
    Note We had to change the output format of the segmentation for the sample. If you obtained a binary segmentation file from this site before 7/1/2010, please upload and segment your video again.
Upload File
Before uploading your video please take the time to review the following points:
  • We support all movie formats that are readable by the current version of ffmpeg.
  • We only have limited computing resources, therefore we restrict videos to an upload size of 50 MB and 50 million voxels, e.g. 350 frames at 500 x 300, 150 frames at 720 x 480, etc.
     However, you will be offered an option to automatically scale and trim the video prior segmentation if it exceeds our limits.
  • Above 50 MB the server will simply reject your file without issueing an error.
  • Use a pixel ratio of 1. We do not pre-scale the video before processing as many video players do, therefore we do not support anamorphic video. This is especially an issue with older MPEG-1 and MPEG-2 data. You will notice that the result-videos look squished along one axis. In this case, scale them manually to have pixel ratio one and upload again.
  • We encourgage you to upload videos with integer fps, otherwise you might encounter a small frame-shift w.r.t. the segmentation result depending on the library you use to read video.
  • Please upload files with a proper file ending, e.g. *.avi, *.mov, *.flv, etc.

File:
Optical flow:
e-mail: optional
Supplying your e-mail address is not necessary. However, if you specify it and the connection breaks down during segmentation, you will be e-mailed the download links containing the segmentation results.
Note: Using optical flow is optional. As reported in our paper, the segmentation quality will improve considerably by the use of optical flow, however there is a computational overhead upfront.
Also note, that the optical flow we use seems to use some random approximations, making the segmentation non-deterministic if used.

Currently 0 segmentation jobs are queued. Your request will be added to the queue and you might encounter a delay until your request can be processed.
Video dimensions are being verified on server. Please wait ...

Note: Please do not reload the web-page during segmentation. Reloading will issue you a new id and you will be inserted into the wait-queue again. Just be patient, it takes some time to compile the results for you, the segmentation runs approximately at 1 fps.

Segmentation is running. Please wait ...
Citation
If you use the segmentation obtained from this website, please reference: 
@article{GrundmannKwatra2010,
   author = {Matthias Grundmann and Vivek Kwatra and Mei Han and Irfan Essa},
   title = {Efficient Hierarchical Graph-Based Video Segmentation},
   journal = {IEEE CVPR},
   year = {2010},
 }
How to process the segmentation?
The segmentation you downloaded is saved as a zipped pb-file, wrapping multiple protocol buffers into one file.
A protobuffer is like an XML container, but offers serialization to binary files.
Download sample program to read and visualize the segmentation protobuffer.

Usage: Press Space to play, 'n' and 'p' advance to next or previous frame respectively, use sliders to random seek to any position.
Copyright

The documents contained in these directories are included by the contributing authors as a means to ensure timely dissemination of scholarly and technical work on a non-commercial basis. Copyright and all rights therein are maintained by the authors or by other copyright holders, notwithstanding that they have offered their works here electronically. It is understood that all persons copying this information will adhere to the terms and constraints invoked by each author's copyright. These works may not be reposted without explicit permission of the copyright holder.