Class Discussion Summary (Jan 26)

Put the citations to papers discussed here:

Panorama generation (Panorama is one of the first widely available computational photography successes)

• Recognizing Panoramas, Brown and Lowe, ICCV 2003.

Scribes: Jiahao Xue, Teresa Jones, Rashmi Chennagiri, Alexander Jules Cardaras, Melanie Wong, Christian Lei

Write the Class Discussion Summary Here:

• Compare and contrast Brown and Lowe's panorama method with ICP for meshes. 
  • e.g. In which ways are they the same algorithm?
  • They both need to choose features/landmarks/points. The Panoramic method chooses features in 2D images but ICP chooses points in 3D model. If you want to use ICP on an image collection, you should find a way to match a large number of feature points, because those are what image collection needs.
  • e.g. What would be needed to get the panorama method to work on 3D meshes?
  • Little to no noise, proper orientation, texture and not point clouds, to add a third dimension to gauge the depth of the points
  • e.g. What would be needed to get ICP to work on image collections?
  • Features in picture to edges to points/vertices and apply the ICP algorithm
• How can we deal with moving objects in panorama creation?
  • e.g. People walking across a courtyard
  • Using machine learning to recognize people, we can separate people’s photo and delete those we don’t need, you can also use the deghosting technique that was discussed in class. This tracks opacity and relates it to the movement, the more opaque, then the more likely they were still.
  • e.g. Panorama of ocean waves
  • By taking long exposure photos, we can deal with the moving waves. Because the water just blurs and this should minimize the influence of moving, some said that it might not be possible. Depends on what aesthetic you are trying to achieve - you could use manual ghosting or you could go for the multi camera techniques to capture movement. We could also try to use the techniques we learned from the first few papers.