simple box structure to capture landmark region
Collective Likelihood: models the average squared distance of images as normal (valid for many pixels, large averages: Central Limit Theorem) for Details see "Markov Chain Monte Carlo for Automated Face Image Analysis" in IJCV 2016
a face detection candidate
evaluate face position with respect to a given detection face box (has scale and position)
Provides specific implementation of a Histogram for a sequence/image of RGB values for RGBA images alpha value is used as mask
evaluate the rendered image with given image evaluator
evaluate each pixel independently, uses alpha channel to determine between foreground and background
evaluate the rendered image with given image evaluator and a segmentation label
evaluate each pixel independently, uses alpha channel to determine between foreground, background and label to destinguish between face and occlusion/non-face.
The LandmarkMapEvaluator evaluates landmark positions by a simple look-up in a LandmarkDetectionMap.
The LandmarkMapEvaluator evaluates landmark positions by a simple look-up in a LandmarkDetectionMap. These maps usually include a noise-model through precomputation.
detection map of this landmark, contains log certainty of detecting landmark at every location
landmarks renderer to calculate landmarks position of current sample
likelihood evaluator for a single landmark position
evaluate position of landmarks with respect to given box returns 0 if all landmarks are in box, else -infinity
evaluate the transformed, rendered image with given image evaluator
Ignore 1-alpha pixels with small likelihood.
Ignore 1-alpha pixels with small likelihood. Evaluate only on alpha-fraction with the largest likelihood.
visualizationCallback: gives you the log likelihood values per pixel. Could be used to visualze the values per pixel.
collection of various pixel color distributions
collection of prior distributions
evaluate each pixel independently, uses alpha channel to determine between foreground, background and label to destinguish between face and occlusion/non-face.
The face region is labeled as 1 (according to the following publication)
This evaluator is an implementation of Equation 2, 4, 5 and 7 of: Occlusion-aware 3D Morphable Face Models, Bernhard Egger, Andreas Schneider, Clemens Blumer, Andreas Morel-Forster, Sandro Schönborn, Thomas Vetter IN: British Machine Vision Conference (BMVC), September 2016 https://dx.doi.org/10.5244/C.30.64