本文整理汇总了Java中gr.iti.mklab.visual.datastructures.IVFPQ类的典型用法代码示例。如果您正苦于以下问题：Java IVFPQ类的具体用法？Java IVFPQ怎么用？Java IVFPQ使用的例子？那么, 这里精选的类代码示例或许可以为您提供帮助。
IVFPQ类属于gr.iti.mklab.visual.datastructures包，在下文中一共展示了IVFPQ类的3个代码示例，这些例子默认根据受欢迎程度排序。您可以为喜欢或者感觉有用的代码点赞，您的评价将有助于系统推荐出更棒的Java代码示例。
示例1: VisualIndex
import gr.iti.mklab.visual.datastructures.IVFPQ; //导入依赖的package包/类
public VisualIndex(String learningFolder, String indexFolder) throws Exception {
int maximumNumVectors = 20000;
int m = 128, k_c = 256, numCoarseCentroids = 8192;
String[] codebookFiles = {
learningFolder + "surf_l2_128c_0.csv",
learningFolder + "surf_l2_128c_1.csv",
learningFolder + "surf_l2_128c_2.csv",
learningFolder + "surf_l2_128c_3.csv"
};
String pcaFile = learningFolder + "pca_surf_4x128_32768to1024.txt";
String coarseQuantizerFile = learningFolder + "qcoarse_1024d_8192k.csv";
String productQuantizerFile = learningFolder + "pq_1024_128x8_rp_ivf_8192k.csv";
ImageVectorization.setFeatureExtractor(new SURFExtractor());
ImageVectorization.setVladAggregator(new VladAggregatorMultipleVocabularies(codebookFiles, numCentroids, AbstractFeatureExtractor.SURFLength));
if (targetLengthMax < initialLength) {
PCA pca = new PCA(targetLengthMax, 1, initialLength, true);
pca.loadPCAFromFile(pcaFile);
ImageVectorization.setPcaProjector(pca);
}
String linearIndexFolder = indexFolder + "/linear";
String ivfpqIndexFolder = indexFolder + "/ivfpq";
this.linearIndex = new Linear(targetLengthMax, maximumNumVectors, false, linearIndexFolder, true, true, 0);
this.ivfpqIndex = new IVFPQ(targetLength, maximumNumVectors, false, ivfpqIndexFolder, m, k_c, PQ.TransformationType.RandomPermutation, numCoarseCentroids, true, 0);
ivfpqIndex.loadCoarseQuantizer(coarseQuantizerFile);
ivfpqIndex.loadProductQuantizer(productQuantizerFile);
ivfpqIndex.setW(128); // how many (out of 8192) lists should be visited during search.
}
示例2: createIndex
import gr.iti.mklab.visual.datastructures.IVFPQ; //导入依赖的package包/类
public void createIndex(String name) throws Exception {
//String ivfpqIndexFolder = "/home/kandreadou/webservice/reveal_indices/" + name + "_" + targetLengthMax;
String ivfpqIndexFolder = "/home/iti-310/VisualIndex/data/"+name+"/ivfpq";
File jeLck = new File(ivfpqIndexFolder, "je.lck");
if (jeLck.exists()) {
jeLck.delete();
}
else{
jeLck.getParentFile().getParentFile().mkdirs();
}
int maximumNumVectors = 100000;
int m2 = 64;
int k_c = 256;
int numCoarseCentroids = 8192;
String coarseQuantizerFile2 = learningFolder + "qcoarse_1024d_8192k.csv";
String productQuantizerFile2 = learningFolder + "pq_1024_64x8_rp_ivf_8192k.csv";
IVFPQ index = new IVFPQ(targetLengthMax, maximumNumVectors, false, ivfpqIndexFolder, m2, k_c, PQ.TransformationType.RandomPermutation, numCoarseCentroids, true, 0);
index.loadCoarseQuantizer(coarseQuantizerFile2);
index.loadProductQuantizer(productQuantizerFile2);
int w = 64; // larger values will improve results/increase seach time
index.setW(w); // how many (out of 8192) lists should be visited during search.
if (indices != null) {
indices.put(name, index);
}
}
示例3: index
import gr.iti.mklab.visual.datastructures.IVFPQ; //导入依赖的package包/类
public static void index(String directory, String linearIndexFolder, String ivfpqIndexFolder, String learningFolder) throws Exception {
File dir = new File(directory);
if(!dir.isDirectory())
return;
if(!learningFolder.endsWith("/"))
learningFolder += "/";
// parameters
int maxNumPixels = 768 * 512;
int[] numCentroids = { 128, 128, 128, 128 };
int initialLength = numCentroids.length * numCentroids[0] * AbstractFeatureExtractor.SURFLength;
int targetLengthMax = 1024;
int targetLength = 1024;
int maximumNumVectors = 20000;
int m = 128, k_c = 256, numCoarseCentroids = 8192;
String[] codebookFiles = {
learningFolder + "surf_l2_128c_0.csv",
learningFolder + "surf_l2_128c_1.csv",
learningFolder + "surf_l2_128c_2.csv",
learningFolder + "surf_l2_128c_3.csv"
};
String pcaFile = learningFolder + "pca_surf_4x128_32768to1024.txt";
String coarseQuantizerFile = learningFolder + "qcoarse_1024d_8192k.csv";
String productQuantizerFile = learningFolder + "pq_1024_128x8_rp_ivf_8192k.csv";
ImageVectorization.setFeatureExtractor(new SURFExtractor());
ImageVectorization.setVladAggregator(new VladAggregatorMultipleVocabularies(codebookFiles,
numCentroids, AbstractFeatureExtractor.SURFLength));
if (targetLengthMax < initialLength) {
PCA pca = new PCA(targetLengthMax, 1, initialLength, true);
pca.loadPCAFromFile(pcaFile);
ImageVectorization.setPcaProjector(pca);
}
Linear linear = new Linear(targetLengthMax, maximumNumVectors, false, linearIndexFolder, true, true, 0);
IVFPQ ivfpq_1 = new IVFPQ(targetLength, maximumNumVectors, false, ivfpqIndexFolder, m, k_c, PQ.TransformationType.RandomPermutation, numCoarseCentroids, true, 0);
ivfpq_1.loadCoarseQuantizer(coarseQuantizerFile);
ivfpq_1.loadProductQuantizer(productQuantizerFile);
ivfpq_1.setW(128); // how many (out of 8192) lists should be visited during search.
for(String id : dir.list()) {
System.out.println(id);
ImageVectorization imvec = new ImageVectorization(dir.toString()+"/", id, targetLengthMax, maxNumPixels);
ImageVectorizationResult imvr = imvec.call();
double[] vector = imvr.getImageVector();
// the full vector is indexed in the disk-based index
linear.indexVector(id, vector);
// the vector is truncated to the correct dimension and renormalized before sending to the ram-based index
double[] newVector = Arrays.copyOf(vector, targetLength);
if (newVector.length < vector.length) {
Normalization.normalizeL2(newVector);
}
ivfpq_1.indexVector(id, newVector);
}
}