1. Illumination and Specularity
  2. Deformation Invariant Matching. Also see for scale-cascaded matching
  3. Multiscale Local Features & Multiscale Global Features
  4. Hybrid Contexts: Eulerian & Lagrangian Shape-Contexts
  5. Randomized Representations -- In the 2008 JAE paper, we perturb patch geometries to account for distributional shifts between the provided data and reality. The initial perturbation models are generalized to deformations learned from scale-cascaded alignment. 
  6. Hierarchical Retrieval
  7. Boosted Retrieval
  8. Relevance Feedback and Crowd-soucring Relevance Feedback
  9. Learning to seek Sparse Human Input Representations for Retrieval