Mercurial > hg > camir-aes2014

annotate core/magnatagatune/MTTTagFeatureGenreBasic.m @ 0:e9a9cd732c1e tip

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
first hg version after svn
author wolffd
date Tue, 10 Feb 2015 15:05:51 +0000
parents
children
rev   line source
wolffd@0 1 classdef MTTTagFeatureGenreBasic < MTTAudioFeature & handle
wolffd@0 2
wolffd@0 3
wolffd@0 4 properties(Constant = true)
wolffd@0 5
wolffd@0 6 my_revision = str2double(substr('$Rev: 99 $', 5, -1));
wolffd@0 7 end
wolffd@0 8
wolffd@0 9 % ---
wolffd@0 10 % here come the internal clip properties.
wolffd@0 11 % the database is stored as a global variable
wolffd@0 12 % ---
wolffd@0 13 properties
wolffd@0 14
wolffd@0 15 my_params = struct(...
wolffd@0 16 ... % ---
wolffd@0 17 ... % these are GenreBasic parameters
wolffd@0 18 ... % ---
wolffd@0 19 'pct_genres', 1, ... % 1/100 percentile genre tags used
wolffd@0 20 'empty_genres', 0 ... % allow empty genres to persist
wolffd@0 21 );
wolffd@0 22 end
wolffd@0 23
wolffd@0 24 methods
wolffd@0 25
wolffd@0 26 % ---
wolffd@0 27 % constructor: pointer to feature in database
wolffd@0 28 % ---
wolffd@0 29 function feature = MTTTagFeatureGenreBasic(varargin)
wolffd@0 30
wolffd@0 31 feature = feature@MTTAudioFeature(varargin{:});
wolffd@0 32 end
wolffd@0 33
wolffd@0 34 function data = extract(feature, clip)
wolffd@0 35 % process tag information and build the tag vector
wolffd@0 36
wolffd@0 37 % ---
wolffd@0 38 % get total number of tags.
wolffd@0 39 %
wolffd@0 40 % NOTE: the highest tag id should match the size of the
wolffd@0 41 % tagdb lexicon
wolffd@0 42 % ---
wolffd@0 43 % num_tags = clip.my_db.genredb.size();
wolffd@0 44
wolffd@0 45 if feature.my_params.pct_genres ~= 0
wolffd@0 46 % no tags, f.e. for parameter experiments in
wolffd@0 47 % higher level features
wolffd@0 48 [tagids, score] = clip.my_db.genredb.annotids_for_owner(clip.id);
wolffd@0 49
wolffd@0 50 % save lexicon
wolffd@0 51 data.lexicon = clip.my_db.genredb.lexicon;
wolffd@0 52
wolffd@0 53 % save to data structure
wolffd@0 54 data.tags.ids = tagids;
wolffd@0 55 data.tags.scores = score;
wolffd@0 56 end
wolffd@0 57
wolffd@0 58 % save info data
wolffd@0 59 data.info.type = 'MTTTagFeatureGenreBasic';
wolffd@0 60 data.info.owner = clip;
wolffd@0 61 data.info.owner_id = clip.id;
wolffd@0 62 data.info.creatorrev = feature.my_revision;
wolffd@0 63
wolffd@0 64 % save parameters
wolffd@0 65 data.info.params = feature.my_params;
wolffd@0 66
wolffd@0 67 data.final.vector = [];
wolffd@0 68 data.final.vector_info = struct();
wolffd@0 69 data.final.dim = 0;
wolffd@0 70 end
wolffd@0 71
wolffd@0 72
wolffd@0 73 % ---
wolffd@0 74 % NOTE: the following transforms are just stated fro
wolffd@0 75 % future use by now
wolffd@0 76 % ---
wolffd@0 77 function define_global_transform(features)
wolffd@0 78 % ---
wolffd@0 79 % compute the relevant tags, and save
wolffd@0 80 % them in the commom place
wolffd@0 81 % ---
wolffd@0 82
wolffd@0 83 % compute extreme cases
wolffd@0 84 if features(1).my_params.pct_genres == 1 && features(1).my_params.empty_genres
wolffd@0 85 % all tags allowed
wolffd@0 86
wolffd@0 87 common.rel_dimensions.tags.ids = 1:numel(features(1).data.lexicon);
wolffd@0 88 common.rel_dimensions.tags.id_pos = 1:numel(features(1).data.lexicon);
wolffd@0 89
wolffd@0 90 % set common feature values
wolffd@0 91 features(1).my_db.set_common(common);
wolffd@0 92 return;
wolffd@0 93
wolffd@0 94 elseif features(1).my_params.pct_genres == 0
wolffd@0 95 % no tags, f.e. for parameter experiments in
wolffd@0 96 % higher level features
wolffd@0 97
wolffd@0 98 common.rel_dimensions.tags.ids = [];
wolffd@0 99 common.rel_dimensions.tags.id_pos = [];
wolffd@0 100
wolffd@0 101 % set common feature values
wolffd@0 102 features(1).my_db.set_common(common);
wolffd@0 103 return;
wolffd@0 104 end
wolffd@0 105
wolffd@0 106
wolffd@0 107 allids = sparse(numel(features), numel(features(1).data.lexicon));
wolffd@0 108 for i = 1:numel(features)
wolffd@0 109
wolffd@0 110 allids(i,features(i).data.tags.ids) = ...
wolffd@0 111 allids(i,features(i).data.tags.ids) + features(i).data.tags.scores;
wolffd@0 112 end
wolffd@0 113
wolffd@0 114 % ---
wolffd@0 115 % get usage of tags and filter not-used tags
wolffd@0 116 % ---
wolffd@0 117 tagsum = sum(allids, 1);
wolffd@0 118 nonzero = tagsum > 0;
wolffd@0 119
wolffd@0 120 % ---
wolffd@0 121 % NOTE: We remove the empty genres, then
wolffd@0 122 % sort by weight / number of appearance and cut off
wolffd@0 123 ids = find(nonzero);
wolffd@0 124 [null, idx] = sort(tagsum(ids),'descend');
wolffd@0 125 ids = ids(idx);
wolffd@0 126
wolffd@0 127 % cutoff
wolffd@0 128 num_tags = ceil( features(1).my_params.pct_genres * numel(ids));
wolffd@0 129 valid_ids = ids(1:min(end, num_tags));
wolffd@0 130
wolffd@0 131 % ---
wolffd@0 132 % NOTE: make sure that the positions for the tags
wolffd@0 133 % stay correctly assigned / consistent
wolffd@0 134 % ---
wolffd@0 135 id_pos = sparse(size(valid_ids));
wolffd@0 136 id_pos(valid_ids) = 1:num_tags;
wolffd@0 137
wolffd@0 138 % save to common data structure
wolffd@0 139 common.rel_dimensions.tags.ids = valid_ids;
wolffd@0 140 common.rel_dimensions.tags.id_pos = id_pos;
wolffd@0 141
wolffd@0 142 % set common feature values
wolffd@0 143 features(1).my_db.set_common(common);
wolffd@0 144 end
wolffd@0 145
wolffd@0 146 function finalise(features)
wolffd@0 147
wolffd@0 148 for i = 1:numel(features)
wolffd@0 149
wolffd@0 150 % check for neccesary parameters
wolffd@0 151 if isempty(features(i).my_db.commondb)
wolffd@0 152
wolffd@0 153 error('Define the global transformation first')
wolffd@0 154 return;
wolffd@0 155 end
wolffd@0 156
wolffd@0 157 if features(i).my_params.pct_genres == 0
wolffd@0 158 vector_info.labels = {};
wolffd@0 159
wolffd@0 160 % save into feature struct
wolffd@0 161 features(i).data.final.dim = 0;
wolffd@0 162 features(i).data.final.vector = [];
wolffd@0 163 features(i).data.final.vector_info = vector_info;
wolffd@0 164 return;
wolffd@0 165 end
wolffd@0 166 % get valid tag ids
wolffd@0 167 valid_ids = features(i).common.rel_dimensions.tags.ids;
wolffd@0 168 num_tags = numel(valid_ids);
wolffd@0 169
wolffd@0 170 % ---
wolffd@0 171 % get vector positions for contined ids,
wolffd@0 172 % and sort out the non_allowed id's
wolffd@0 173 % ---
wolffd@0 174 id_pos = features(i).common.rel_dimensions.tags.id_pos;
wolffd@0 175
wolffd@0 176 [has_ids, has_pos] = intersect(features(i).data.tags.ids, valid_ids);
wolffd@0 177
wolffd@0 178 score = features(i).data.tags.scores;
wolffd@0 179
wolffd@0 180 % create feature vector
wolffd@0 181 vector = zeros(num_tags,1);
wolffd@0 182 vector(id_pos(has_ids)) = score(has_pos);
wolffd@0 183
wolffd@0 184 % ---
wolffd@0 185 % NOTE: this labelling costs to much space
wolffd@0 186 % ---
wolffd@0 187 vector_info.labels = features(i).data.lexicon(valid_ids);
wolffd@0 188
wolffd@0 189 % save into feature struct
wolffd@0 190 features(i).data.final.dim = num_tags;
wolffd@0 191 features(i).data.final.vector = vector;
wolffd@0 192 features(i).data.final.vector_info = vector_info;
wolffd@0 193 end
wolffd@0 194 end
wolffd@0 195
wolffd@0 196 function [a1] = visualise(feature)
wolffd@0 197
wolffd@0 198 % get tag descriptions form first clip
wolffd@0 199 tags = MTTClip(feature(1).owner_id()).my_db.genres;
wolffd@0 200
wolffd@0 201 % works for multiple feature instances
wolffd@0 202 for i = 1:numel(feature)
wolffd@0 203
wolffd@0 204 clip = MTTClip(feature(i).owner_id());
wolffd@0 205
wolffd@0 206 % plot feature data
wolffd@0 207 h = bar(feature(i).data.final.vector);
wolffd@0 208 a1 = gca;
wolffd@0 209
wolffd@0 210 set(a1,'XTick', 1:feature(i).data.final.dim,...
wolffd@0 211 'XTickLabel', tags);
wolffd@0 212 title(sprintf('clip %d: %s by %s, genres', ...
wolffd@0 213 clip.id, clip.title(),clip.artist()));
wolffd@0 214 end
wolffd@0 215 end
wolffd@0 216
wolffd@0 217 end
wolffd@0 218 end