Package viper.comparison

Provides the classes necessary to compare Descriptors, including a simple command line application useful for generating graphs between a large number of cases.

See:
Description

Interface Summary

Evaluation	An evaluation is a a high-level method of comparing two sets of descriptors.
Evaluation.Information	Evaluation.Information class for describing evaluation results

Class Summary

Comparison	This class represents the differences between two Descriptors.
CompEvaluator	A collection of Descriptor Objects that supports various operations on them, such as comparison between two DescVectors, collection of statistics, reading and printing from files, and various others.
CompFilter	Classes that implement this take in a CompMatrix and upgrade all comparisons that meet a certain criterion.
CompMatrix	A collection of Descriptor Objects that supports various operations on them, such as comparison between two DescriptorLists, collection of statistics, reading and printing from files, and various others.
EvaluationParameters	Describes a set of parameters for setting up an evaluation.
FramewiseEvaluation	A framewise evaluation views each frame individually, computing aggregate statisitcs on a frame-by-frame or even pixel-by-pixel level, disregarding the continuity of descriptors.
FramewiseEvaluation.FrameMeasure	A measure for a given frame.
ObjectEvaluation	The standard evaluation that counts the number of matching descriptors.
ObjectInformation	Stores precision/recall information for object evaluation.
TrackingEvaluation	Evaluates keyed descriptor pairs, as described by Kasturi et. al. in their paper.
TrackingEvaluation.TrackingMeasure	A measure specifically designed for tracking, this allows multiple child measures.
TrackingInformation	Collects scores for the tracking evaluatoin type
ViperPE	This class offers a function to compare two Descriptor data files.

Package viper.comparison Description

Provides the classes necessary to compare Descriptors, including a simple command line application useful for generating graphs between a large number of cases.

Two sets of Descriptors can be compared against each other via the CompEvaluator class. This class takes in two sets of Descriptors in list form and, via various methods to perform certain computations, can compare them to determine the overall distance between the two sets.

In order to compare two sets of descriptors from files, the following sequence might be followed (this is pretty much what GtfC does):

1. Read in the Descriptor configuration data.
2. Read in the Equivalencies.
3. Read in the Evaluations and Limitations.
4. Read in the Descriptor data.
5. Initialize the CompMatrix.
6. Cycle through the comparison functions to the desired accuracy of distance measure. After each, print the results if wanted. Remember, Cropping may be performed at any time, but the rest must be performed in order.
   1. Detection
   2. Localization
   3. Statistical Match
   4. Cropping
7. Display results, either in the standard format or by parsing the raw format into a graph.

Standard Output Format

The format prints out the results in a readable format. It's format is as follows:

• Input Parameters
  • The different filenames read in. The properties file, data files, etc.
  • The level and match type, eg level 3 (Statistics) and Multiple type matching
  • The statistic and tolerance used for thresholding.
• Filters
• Metrics and tolerances for different descriptor types and their attributes
• False and Missed Descriptors
  Depending on the level, these may include:
  • Level 1: Detection. Do some some frames of a target/candidate pair overlap with similar attributes?
  • Level 2: Localization. Do enough of the frames contain matching data?
  • Level 3: Statistics. Does some statistic (Mean, Median, Max) of the frame distances meet some threshold?
  • Cropping: What is the best match for given targets? What set of pairs minimizes overall distance?
• Detections. What targets get what candidates and how close the match is.
• Summary Results. The overall precision and recall.
• Variable Attribute Results.

See Also:

the package defining Descriptors

, the package defining Attributes