Package org.hd.d.pg2k.ai.scorer

AI code for predicting the popularity or "score" of still images.

See:
          Description

Interface Summary
AbstractImgScorer.ARGBPixelFilter Interface to accept or reject putative samples in getSamplePoints().
BadScorer Empty marker interface implemented by test/dummy/bad Scorers.
ScorerCacheIF Base interface to compute (and cache) the score and confidence for exhibits.
ScorerIF Base interface to compute the score and confidence for an exhibit.
ScorerParam  
ScorerPopulation.NewBestCallbackIF Interface of call-back hook to report a new "best" Scorer by name, eg to the system event record.
 

Class Summary
AbstractImgSampleScorer Base interface to compute the score and confidence for a 2D still image using pixel ARGB sampling.
AbstractImgScorer Base interface to compute the score and confidence for a 2D still image.
AbstractNonParamScorer Simple abstract super class for some non-parameterised Scorers.
AbstractScorer Class to implement some common Scorer features in one place.
AbstractScorerCache Shared abstract base to handle common Scorer cache tasks.
MiniScorerCacheImpl "Lite" implementation to compute (and cache) the score and confidence for exhibits.
ScoreAndConf Immutable (and Internable) return type from scorer method.
ScorerCacheImpl Simple/default implementation to compute (and cache) the score and confidence for exhibits.
ScorerCreator Creates new Scorer (parameter-set) instances given an existing population.
ScorerCreator.ScorerWork Class to encapsulate all background and evolution work for a given ScorerCache.
ScorerParamEnum<E extends Enum<E>> Immutable (enum) Scorer parameter.
ScorerParamInteger Immutable (int) Scorer parameter.
ScorerPopulation Thread-safe container for Scorer-and-parameter information to maintain an evolving population.
 

Package org.hd.d.pg2k.ai.scorer Description

AI code for predicting the popularity or "score" of still images.

The aim of this code is, given a still image as a RenderedImage, to compute a predicted popularity (0--100%) and a condfidence (0.0--0.1).

The aim is to be reasonably accurate yet use little memory or CPU.

The predictor code is shipped as a "safe" Java class that can run in a very limited sandbox (eg for an Applet of JWS) and that is guaranteed to run in finite time and with limited resources. This class complies with a simple standard interface, and thus can be plugged back into the live server or run in the the user's browser or (for example) the JWS upload tool.

This system can be calibrated/trained against the live vote database.


DHD Multimedia Gallery V1.60.69

Copyright (c) 1996-2012, Damon Hart-Davis. All rights reserved.