The Idea
Measure overlap in hyperspace
The Idea
Compute Specificity and Generalisation ability
The Idea
Repeat for all cross-pairings
Measuring Distance
Use the shuffle transform
Measuring Distance
Shuffle: face example
Measuring Distance
Shuffle: face example
Shuffle: brain example
Measuring Distance
Shuffle: brain example
Measuring Distance
Distance matrix for the sets
Models - Face
Input images
Models - Face
Poor output image
Models - Face
Models changed by perturbation of landmark points
Models - Face
Models changed by large deformation of training images
Models - Face
Models changed by smaller deformation of training images
Models - Face
The correct model (Surrey database)
Models - Brain
Model automatically built by group-wise registration
Models - Brain
Warped models (CPS)
Radically-warped models (GIMP)
Models - Brain
Sequence of warps and difference images (bottom)
Evaluation of models - Faces
Generalisation ability as landmark points are shifted
Evaluation of models - Faces
Specificity as landmark points are shifted
Evaluation of models - Brains
Generalisation ability as landmark points are shifted
Evaluation of models - Brains
Specificity as landmark points are shifted
Evaluation of models - Brains
Specificity as images in the training set are deformed
Comparison - Brain Model Construction
Generalisation ability for the different approaches
Comparison - Brain Model Construction
In more depth
Comparison - Brain Model Construction
Specificity for the different approaches
Comparison - Brain Model Construction
In more depth
Comparing Different Shuffle Distances
Generalisation ability for the different window sizes
Comparing Different Shuffle Distances
Specificity for the different window sizes
Normalisation - Framework
The method of normalising (perhaps not practically useful)
Normalisation
Generalisation ability for the original training set (brain)
Normalisation
Generalisation ability for the pseudo training set
Normalisation
Normalised Generalisation ability
Normalisation
Normalised Specificity