The Semantic Robot Vision Challenge

NEWS

• Meetings are Fridays at, 11 am, in icics 146

Some Stuff

• The Rules page at CMU

• Dave Meger's whiteboard transcription subTasks.pdf

• Sequence from Bart right.mpg

• Stereo processing of sequence stereo.mpg

• Sample images from 1.3m height cat01.png heli03.png keg01.png robo03.png

• A data-set of 31 views of objects captured from Bart.
  • look under /lci/project/raid1/srvc/bumblebee1
  • frameXX_depth0.png Raw disparity image from PGR library
  • frameXX_depth.png Scaled disparity image to look at
  • frameXX_left.png Left frame 1024x768 RGB
  • frameXX_right.png Right frame 1024x798 RGB
  • Here is a composite showing these four images: bumblebee1_example.png

• Canon capture library at sourceforge: http://capture.sourceforge.net/
  • I've (Scott) installed the software and tried using it with my Canon A70. I haven't tried all the features, but it was fairly straightforward to zoom, use the flash, and capture and download a photo. Not wickedly fast though.

• An unintelligent Google image search shell script: fetcher.sh
  • Usage: ./fetcher "search string"
  • Examples:
    • To search for tiger: ./fetcher "tiger"
    • To search for tiger and woods: ./fetcher "tiger woods"
    • To search for the exact phrase "tiger woods": ./fetcher "\"tiger woods\""
    • To search for tiger with no mention of woods: ./fetcher "tiger -woods"
  • The script will create a folder named similarly to the search string and place the images returned on the first page of Google's image search into the folder. If two images have the same name, the first one will be saved normally (i.e. image.jpg), but the next one will be saved with an extra number (i.e. image.jpg.1).

• I have set up a Subversion (SVN) repository which can act as a central location for our project code.
  • The repository is located at /lci/project/raid1/srvc/SVN/.
  • Only members of the Unix group srvc can access this location. Many of the team are already members of this group. Please contact me (Dave Meger) or mail help@csDELETEthisTEXT.ubc.ca if you'd like to be added.
  • During our initial development, I have proposed separating the repository based on the rough structure our teams seem to be taking on. Please add your code under the appropriate folder, and organize the lower-levels as you see fit. This can easily be changed in the future, but currently the top-level directories are:
    • CLASSIFY
    • DOC
    • ROBOT
    • WEB
  • To get started, look at a brief tutorial: SRVC SVN Tutorial. Also, the Subversion book listed below is a good reference for those getting started with SVN. If you require a more detailed explanation, ask a team member, or we can run through a quick demo in a weekly meeting.

Organization stuff

• Version Control with Subversion - A free online book about using subversion
• Doxygen documentation system

Image libraries

• Adobe Generic Image Library (GIL)
  • Accepted for inclusion in the Boost C++ Library
  • Write generic algorithms that will work for any image type
  • Nice online presentation by the lead developer here: GIL Presentation
  • Library does not include pre-written algorithms like OpenCV, or VXL
• VXL
• VIGRA
  • The original generic image processing library
  • See Reusable software in computer vision and STL-style generic programming with images for the original motivation for generic image libraries.
  • Much like GIL, but without support from something like Adobe
  • Does include collection of low-level processing algorithms

Comments

Minutes

• February 16 pdf

Team Members

• Scott Helmer
• Per-Erik Forssen
• Simon Leonard
• Julia Vogel
• Kenji Okuma
• Pooyan Fazli
• Ara
• Dave Meger
• Sancho Mccann
• Marius

Feel free to add anything related to the challenge here.

-- PerErikForssen - 07 Feb 2007