private void indexFiles(String dir, String index, int featureIndex, boolean createNewIndex)
throws IOException {
ArrayList<String> images = FileUtils.getAllImages(new File(dir), true);
IndexWriter iw =
LuceneUtils.createIndexWriter(
index, createNewIndex, LuceneUtils.AnalyzerType.WhitespaceAnalyzer);
// select one feature for the large index:
int count = 0;
long ms = System.currentTimeMillis();
DocumentBuilder builder = new ChainedDocumentBuilder();
((ChainedDocumentBuilder) builder).addBuilder(builders[featureIndex]);
// ((ChainedDocumentBuilder) builder).addBuilder(builders[0]);
for (Iterator<String> iterator = images.iterator(); iterator.hasNext(); ) {
count++;
if (count > 100 && count % 5000 == 0) {
System.out.println(
count
+ " files indexed. "
+ (System.currentTimeMillis() - ms) / (count)
+ " ms per file");
}
String file = iterator.next();
try {
iw.addDocument(builder.createDocument(new FileInputStream(file), file));
} catch (Exception e) {
System.err.println("Error: " + e.getMessage());
}
}
iw.close();
}
public void testSearchRunTime() throws IOException {
int queryDocID;
IndexReader reader = DirectoryReader.open(FSDirectory.open(new File("index-large-new")));
int featureIndex = 0;
ImageSearchHits hits = searchers[featureIndex].search(reader.document(0), reader);
hits = searchers[featureIndex].search(reader.document(1), reader);
long ms = System.currentTimeMillis();
for (int i = 0; i < 100; i++) {
queryDocID = i;
// select one feature for the large index:
hits = searchers[featureIndex].search(reader.document(queryDocID), reader);
}
ms = System.currentTimeMillis() - ms;
System.out.println("ms = " + ms / 100);
}
public void testRerankFilters() throws IOException {
int queryDocID = (int) (Math.random() * 10000);
IndexReader reader = DirectoryReader.open(FSDirectory.open(new File("index-large")));
// select one feature for the large index:
int featureIndex = 4;
int count = 0;
long ms = System.currentTimeMillis();
ImageSearchHits hits = searchers[featureIndex].search(reader.document(queryDocID), reader);
RerankFilter rerank = new RerankFilter(featureClasses[0], DocumentBuilder.FIELD_NAME_CEDD);
LsaFilter lsa = new LsaFilter(featureClasses[0], DocumentBuilder.FIELD_NAME_CEDD);
FileUtils.saveImageResultsToPng(
"GeneralTest_rerank_0_old",
hits,
reader.document(queryDocID).getValues(DocumentBuilder.FIELD_NAME_IDENTIFIER)[0]);
hits = rerank.filter(hits, reader.document(queryDocID));
FileUtils.saveImageResultsToPng(
"GeneralTest_rerank_1_new",
hits,
reader.document(queryDocID).getValues(DocumentBuilder.FIELD_NAME_IDENTIFIER)[0]);
hits = lsa.filter(hits, reader.document(queryDocID));
FileUtils.saveImageResultsToPng(
"GeneralTest_rerank_2_lsa",
hits,
reader.document(queryDocID).getValues(DocumentBuilder.FIELD_NAME_IDENTIFIER)[0]);
}
public void testIndexLarge() throws IOException {
// ArrayList<String> images = FileUtils.getAllImages(new
// File("C:\\Temp\\testImagelogos"), true);
ArrayList<String> images =
FileUtils.getAllImages(
new File("C:\\Java\\Projects\\LireSVN\\testdata\\flickr-10000"), false);
IndexWriter iw =
LuceneUtils.createIndexWriter(
"index-large", true, LuceneUtils.AnalyzerType.WhitespaceAnalyzer);
// select one feature for the large index:
int featureIndex = 13;
int count = 0;
long ms = System.currentTimeMillis();
DocumentBuilder builder = new ChainedDocumentBuilder();
((ChainedDocumentBuilder) builder).addBuilder(builders[featureIndex]);
// ((ChainedDocumentBuilder) builder).addBuilder(builders[0]);
for (Iterator<String> iterator = images.iterator(); iterator.hasNext(); ) {
count++;
if (count > 100 && count % 500 == 0) {
System.out.println(
count
+ " files indexed. "
+ (System.currentTimeMillis() - ms) / (count)
+ " ms per file");
}
String file = iterator.next();
try {
// try to trim the image first ....
// BufferedImage img = ImageUtils.trimWhiteSpace(ImageIO.read(new
// FileInputStream(file)));
// iw.addDocument(builder.createDocument(img, file));
iw.addDocument(builder.createDocument(new FileInputStream(file), file));
} catch (Exception e) {
e
.printStackTrace(); // To change body of catch statement use File | Settings | File
// Templates.
}
}
iw.close();
}
private void indexFiles(ArrayList<String> images, DocumentBuilder builder, String indexPath)
throws IOException {
// System.out.println(">> Indexing " + images.size() + " files.");
// DocumentBuilder builder = DocumentBuilderFactory.getExtensiveDocumentBuilder();
// DocumentBuilder builder = DocumentBuilderFactory.getFastDocumentBuilder();
IndexWriter iw = LuceneUtils.createIndexWriter(indexPath, true);
int count = 0;
long time = System.currentTimeMillis();
for (String identifier : images) {
Document doc = builder.createDocument(new FileInputStream(identifier), identifier);
iw.addDocument(doc);
count++;
if (count % 100 == 0) System.out.println(count + " files indexed.");
// if (count == 200) break;
}
long timeTaken = (System.currentTimeMillis() - time);
float sec = ((float) timeTaken) / 1000f;
System.out.println(sec + " seconds taken, " + (timeTaken / count) + " ms per image.");
iw.commit();
iw.close();
}
public void testReUse() throws IOException, IllegalAccessException, InstantiationException {
ArrayList<String> testFiles = FileUtils.getAllImages(new File("testdata/ferrari"), true);
for (Class c : featureClasses) {
LireFeature f1 = (LireFeature) c.newInstance();
System.out.println(c.getName());
for (String testFile : testFiles) {
f1.extract(ImageIO.read(new File(testFile)));
LireFeature f2 = (LireFeature) c.newInstance();
f2.extract(ImageIO.read(new File(testFile)));
// System.out.println(Arrays.toString(f1.getDoubleHistogram()));
// System.out.println(Arrays.toString(f2.getDoubleHistogram()));
assertEquals(f2.getDistance(f1), 0d, 0.000000001);
f2.setByteArrayRepresentation(f1.getByteArrayRepresentation());
assertEquals(f2.getDistance(f1), 0d, 0.000000001);
byte[] tmp = new byte[1024 * 100];
Arrays.fill(tmp, (byte) 0x000F);
byte[] bytes = f1.getByteArrayRepresentation();
System.arraycopy(bytes, 0, tmp, 12, bytes.length);
f2.setByteArrayRepresentation(tmp, 12, bytes.length);
assertEquals(f2.getDistance(f1), 0d, 0.000000001);
}
}
}
public void testSearchIndexLarge() throws IOException {
for (int i = 0; i < 10; i++) {
int queryDocID = (int) (Math.random() * 800);
// queryDocID = 877 * (i + 1);
IndexReader reader = DirectoryReader.open(FSDirectory.open(new File("index-large")));
// select one feature for the large index:
int featureIndex = 13;
int count = 0;
long ms = System.currentTimeMillis();
ImageSearchHits hits = searchers[featureIndex].search(reader.document(queryDocID), reader);
for (int j = 0; j < hits.length(); j++) {
String fileName = hits.doc(j).getValues(DocumentBuilder.FIELD_NAME_IDENTIFIER)[0];
System.out.println(hits.score(j) + ": \t" + fileName);
}
// FileUtils.saveImageResultsToHtml("GeneralTest_testSearchIndexLarge_", hits,
// reader.document(10).getValues(DocumentBuilder.FIELD_NAME_IDENTIFIER)[0]);
FileUtils.saveImageResultsToPng(
"GeneralTest_testSearchIndexLarge_" + i + "_",
hits,
reader.document(queryDocID).getValues(DocumentBuilder.FIELD_NAME_IDENTIFIER)[0]);
}
}
public void testClassify() throws IOException {
boolean weightByRank = true;
String[] classes = {
"2012", "beach", "food", "london", "music", "nature", "people", "sky", "travel", "wedding"
};
int k = 50;
// CONFIG
String fieldName = DocumentBuilder.FIELD_NAME_COLORLAYOUT;
LireFeature feature = new ColorLayout();
String indexPath = "E:\\acmgc-cl-idx";
System.out.println(
"Tests for feature "
+ fieldName
+ " with k="
+ k
+ " - weighting by rank sum: "
+ weightByRank);
System.out.println("========================================");
HashMap<String, Integer> tag2count = new HashMap<String, Integer>(k);
HashMap<String, Double> tag2weight = new HashMap<String, Double>(k);
int c = 9; // used for just one class ...
// for (int c = 0; c < 10; c++) {
String classIdentifier = classes[c];
String listFiles = "D:\\DataSets\\Yahoo-GC\\test\\" + classIdentifier + ".txt";
// INIT
int[] confusion = new int[10];
Arrays.fill(confusion, 0);
HashMap<String, Integer> class2id = new HashMap<String, Integer>(10);
for (int i = 0; i < classes.length; i++) class2id.put(classes[i], i);
BufferedReader br = new BufferedReader(new FileReader(listFiles));
String line;
IndexReader ir = DirectoryReader.open(MMapDirectory.open(new File(indexPath)));
// in-memory linear search
// ImageSearcher bis = new GenericFastImageSearcher(k, feature.getClass(), fieldName,
// true, ir);
// hashing based searcher
BitSamplingImageSearcher bis =
new BitSamplingImageSearcher(k, fieldName, fieldName + "_hash", feature, 1000);
ImageSearchHits hits;
int count = 0, countCorrect = 0;
long ms = System.currentTimeMillis();
while ((line = br.readLine()) != null) {
try {
tag2count.clear();
tag2weight.clear();
hits = bis.search(ImageIO.read(new File(line)), ir);
// set tag weights and counts.
for (int l = 0; l < k; l++) {
String tag = getTag(hits.doc(l));
if (tag2count.get(tag) == null) tag2count.put(tag, 1);
else tag2count.put(tag, tag2count.get(tag) + 1);
if (weightByRank) {
if (tag2weight.get(tag) == null) tag2weight.put(tag, (double) l);
else tag2weight.put(tag, (double) l + tag2weight.get(tag));
} else {
if (tag2weight.get(tag) == null) tag2weight.put(tag, Double.valueOf(hits.score(l)));
else tag2weight.put(tag, (double) l + hits.score(l));
}
}
// find class:
int maxCount = 0, maxima = 0;
String classifiedAs = null;
for (Iterator<String> tagIterator = tag2count.keySet().iterator();
tagIterator.hasNext(); ) {
String tag = tagIterator.next();
if (tag2count.get(tag) > maxCount) {
maxCount = tag2count.get(tag);
maxima = 1;
classifiedAs = tag;
} else if (tag2count.get(tag) == maxCount) {
maxima++;
}
}
// if there are two or more classes with the same number of results, then we take a look at
// the weights.
// else the class is alread given in classifiedAs.
if (maxima > 1) {
double minWeight = Double.MAX_VALUE;
for (Iterator<String> tagIterator = tag2count.keySet().iterator();
tagIterator.hasNext(); ) {
String tag = tagIterator.next();
if (tag2weight.get(tag) < minWeight) {
minWeight = tag2weight.get(tag);
classifiedAs = tag;
}
}
}
// if (tag2.equals(tag3)) tag1 = tag2;
count++;
if (classifiedAs.equals(classIdentifier)) countCorrect++;
// confusion:
confusion[class2id.get(classifiedAs)]++;
// System.out.printf("%10s (%4.3f, %10d, %4d)\n", classifiedAs, ((double)
// countCorrect / (double) count), count, (System.currentTimeMillis() - ms) / count);
} catch (Exception e) {
System.err.println(e.getMessage());
}
}
// System.out.println("Results for class " + classIdentifier);
System.out.printf("Class\tAvg. Precision\tCount Test Images\tms per test\n");
System.out.printf(
"%s\t%4.5f\t%10d\t%4d\n",
classIdentifier,
((double) countCorrect / (double) count),
count,
(System.currentTimeMillis() - ms) / count);
System.out.printf("Confusion\t");
// for (int i = 0; i < classes.length; i++) {
// System.out.printf("%s\t", classes[i]);
// }
// System.out.println();
for (int i = 0; i < classes.length; i++) {
System.out.printf("%d\t", confusion[i]);
}
System.out.println();
// }
}
protected void setUp() throws Exception {
super.setUp();
// set to all queries ... approach "leave one out"
sampleQueries = new int[1000];
for (int i = 0; i < sampleQueries.length; i++) {
sampleQueries[i] = i;
}
indexPath += "-" + System.currentTimeMillis() % (1000 * 60 * 60 * 24 * 7);
// Setting up DocumentBuilder:
// parallelIndexer = new ParallelIndexer(8, indexPath, testExtensive);
parallelIndexer =
new ParallelIndexer(8, indexPath, testExtensive, true) {
@Override
public void addBuilders(ChainedDocumentBuilder builder) {
// builder.addBuilder(DocumentBuilderFactory.getCEDDDocumentBuilder());
//
// builder.addBuilder(DocumentBuilderFactory.getAutoColorCorrelogramDocumentBuilder());
builder.addBuilder(DocumentBuilderFactory.getColorLayoutBuilder());
builder.addBuilder(DocumentBuilderFactory.getEdgeHistogramBuilder());
// builder.addBuilder(DocumentBuilderFactory.getFCTHDocumentBuilder());
builder.addBuilder(DocumentBuilderFactory.getJCDDocumentBuilder());
//
// builder.addBuilder(DocumentBuilderFactory.getJointHistogramDocumentBuilder());
builder.addBuilder(DocumentBuilderFactory.getOpponentHistogramDocumentBuilder());
builder.addBuilder(DocumentBuilderFactory.getPHOGDocumentBuilder());
//
// builder.addBuilder(DocumentBuilderFactory.getColorHistogramDocumentBuilder());
builder.addBuilder(DocumentBuilderFactory.getScalableColorBuilder());
//
// builder.addBuilder(DocumentBuilderFactory.getLuminanceLayoutDocumentBuilder());
//
// builder.addBuilder(DocumentBuilderFactory.getJpegCoefficientHistogramDocumentBuilder());
//
// builder.addBuilder(DocumentBuilderFactory.getColorHistogramDocumentBuilder());
// builder.addBuilder(DocumentBuilderFactory.getGaborDocumentBuilder());
// builder.addBuilder(DocumentBuilderFactory.getTamuraDocumentBuilder());
// builder.addBuilder(DocumentBuilderFactory.getScalableColorBuilder());
// builder.addBuilder(new GenericDocumentBuilder(RankAndOpponent.class,
// "jop"));
// builder.addBuilder(new
// GenericFastDocumentBuilder(FuzzyOpponentHistogram.class, "opHist"));
// builder.addBuilder(new SurfDocumentBuilder());
// builder.addBuilder(new MSERDocumentBuilder());
// builder.addBuilder(new SiftDocumentBuilder());
// builder.addBuilder(new GenericDocumentBuilder(SPCEDD.class,
// "spcedd"));
// builder.addBuilder(new GenericDocumentBuilder(SPFCTH.class,
// "spfcth"));
// builder.addBuilder(new GenericDocumentBuilder(SPJCD.class, "spjcd"));
// builder.addBuilder(new GenericDocumentBuilder(SPACC.class, "spacc"));
// builder.addBuilder(new
// GenericDocumentBuilder(LocalBinaryPatterns.class, "lbp"));
// builder.addBuilder(new
// GenericDocumentBuilder(BinaryPatternsPyramid.class, "whog"));
// builder.addBuilder(new
// GenericDocumentBuilder(LocalBinaryPatternsAndOpponent.class, "jhl"));
// builder.addBuilder(new
// GenericDocumentBuilder(RotationInvariantLocalBinaryPatterns.class, "rlbp"));
// builder.addBuilder(new GenericDocumentBuilder(SPLBP.class, "splbp"));
}
};
}
public void computeMAP(ImageSearcher searcher, String prefix, IndexReader reader)
throws IOException {
Pattern p = Pattern.compile("([0-9]+).jpg");
double map = 0;
double errorRate = 0d;
double precision10 = 0d;
double[] pr10cat = new double[10];
double[] pr10cnt = new double[10];
for (int i = 0; i < pr10cat.length; i++) {
pr10cat[i] = 0d;
pr10cnt[i] = 0d;
}
long sum = 0, ms = 0;
for (int i = 0; i < sampleQueries.length; i++) {
int id = sampleQueries[i];
String file = testExtensive + "/" + id + ".jpg";
ms = System.currentTimeMillis();
ImageSearchHits hits = searcher.search(findDoc(reader, id + ".jpg"), reader);
sum += (System.currentTimeMillis() - ms);
int goodOnes = 0;
double avgPrecision = 0d;
double precision10temp = 0d;
int countResults = 0;
for (int j = 0; j < hits.length(); j++) {
Document d = hits.doc(j);
String hitsId = d.getValues(DocumentBuilder.FIELD_NAME_IDENTIFIER)[0];
Matcher matcher = p.matcher(hitsId);
if (matcher.find()) hitsId = matcher.group(1);
else fail("Did not get the number ...");
int testID = Integer.parseInt(hitsId);
if (testID != id) countResults++;
if ((testID != id) && ((int) Math.floor(id / 100) == (int) Math.floor(testID / 100))) {
goodOnes++;
// Only if there is a change in recall
avgPrecision += (double) goodOnes / (double) countResults;
// System.out.print("x");
if (j <= 10) {
precision10temp += 1d;
}
} else {
if (j == 1) { // error rate
errorRate++;
}
}
} // end for loop iterating results.
// if (avgPrecision<=0) {
// System.out.println("avgPrecision = " + avgPrecision);
// System.out.println("goodOnes = " + goodOnes);
// }
assertTrue("Check if average precision is > 0", avgPrecision > 0);
assertTrue("Check if goodOnes is > 0", goodOnes > 0);
avgPrecision = avgPrecision / goodOnes;
precision10 += precision10temp / 10d;
// precision @ 10 for each category ...
pr10cat[(int) Math.floor(id / 100)] += precision10temp / 10d;
pr10cnt[(int) Math.floor(id / 100)] += 1d;
map += avgPrecision;
}
map = map / sampleQueries.length;
errorRate = errorRate / sampleQueries.length;
precision10 = precision10 / sampleQueries.length;
System.out.print(prefix + "\t");
System.out.print(String.format("%.5f", map) + '\t');
System.out.print(String.format("%.5f", precision10) + '\t');
System.out.print(String.format("%.5f", errorRate) + '\t');
// precision@10 per category
for (int i = 0; i < pr10cat.length; i++) {
double v = 0;
if (pr10cnt[i] > 0) v = pr10cat[i] / pr10cnt[i];
// System.out.print(i + ": ");
System.out.printf("%.5f\t", v);
}
System.out.printf("%2.3f\t", (double) sum / (double) sampleQueries.length);
System.out.println();
}