This is an archive of image data sets used to test the application of non-linear dimensionality reduction methods (such as Isomap and LLE) to images. This site is under construction, please e-mail me if you find this useful or have suggestions.
Image manifolds
  1. A set views of an small statue from different viewpoints.
    fileename convention: "robert_xxx_00yyy.jpg" where xxx is the camera elevation angle (every 6 degrees from 6-84), and yyy is the object rotation angle (every 6 degrees from 0-354).
    Paper Boy statue data (60.9 MBytes, zip file)
    The sparse data set contains the same images but sampled every 12 degrees.
    Sparse Paper Boy statue data (13.4 MBytes, zip file). This data replicates the data used in the paper "Using Thousands of Images of an Object" (bib|(PDF, 330 kb))
  2. Several clips of birds in flight. Birds in flight are naturally represented as a 2-D manifold with 1 cyclic dimension (phase of the bird flap) and 1 linear dimension (size of the bird on the image).
    1. A bird flying over Forest Park in St. Louis Missouri, (10 Mbytes).
    2. Another bird switching from flying to gliding (8.9 Mbytes).
  3. Image data with less clear topological structure: "Charlie Chaplin" video (avi).
  4. Image datasets used in the original Isomap paper (link).
  5. The page of Piotr Dollar, containing facial expression datasets that are examples of (perhaps 1-D) image manifolds. (link).
  6. Datasets from the Semi-Definate Embedding paper (link).
This page supported in part by NSF IIS-0413291, "Non-Parametric Representations of Motions and Actions in Video".

Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.
back to the WuMap homepage