Java OpenImaj-识别形状
我的目的是在训练了一个分类器之后识别一个形状，类似于OpenIMAJ教程第12章中所做的。 第12章使用Caltech101类，这对我没有帮助，因为我想使用自己的一组图像来训练分类器。我根据第12章创建了此工作代码：Java OpenImaj-识别形状,java,openimaj,Java,Openimaj,我的目的是在训练了一个分类器之后识别一个形状，类似于OpenIMAJ教程第12章中所做的。 第12章使用Caltech101类，这对我没有帮助，因为我想使用自己的一组图像来训练分类器。我根据第12章创建了此工作代码： package com.mycompany.video.analytics; import de.bwaldvogel.liblinear.SolverType; import org.openimaj.data.DataSource; import org.openimaj.d
package com.mycompany.video.analytics;
import de.bwaldvogel.liblinear.SolverType;
import org.openimaj.data.DataSource;
import org.openimaj.data.dataset.Dataset;
import org.openimaj.data.dataset.GroupedDataset;
import org.openimaj.data.dataset.ListDataset;
import org.openimaj.data.dataset.VFSGroupDataset;
import org.openimaj.experiment.dataset.sampling.GroupSampler;
import org.openimaj.experiment.dataset.sampling.GroupedUniformRandomisedSampler;
import org.openimaj.experiment.dataset.split.GroupedRandomSplitter;
import org.openimaj.experiment.evaluation.classification.ClassificationEvaluator;
import org.openimaj.experiment.evaluation.classification.ClassificationResult;
import org.openimaj.experiment.evaluation.classification.analysers.confusionmatrix.CMAnalyser;
import org.openimaj.experiment.evaluation.classification.analysers.confusionmatrix.CMResult;
import org.openimaj.feature.DoubleFV;
import org.openimaj.feature.FeatureExtractor;
import org.openimaj.feature.SparseIntFV;
import org.openimaj.feature.local.data.LocalFeatureListDataSource;
import org.openimaj.feature.local.list.LocalFeatureList;
import org.openimaj.image.FImage;
import org.openimaj.image.ImageUtilities;
import org.openimaj.image.feature.dense.gradient.dsift.ByteDSIFTKeypoint;
import org.openimaj.image.feature.dense.gradient.dsift.DenseSIFT;
import org.openimaj.image.feature.dense.gradient.dsift.PyramidDenseSIFT;
import org.openimaj.image.feature.local.aggregate.BagOfVisualWords;
import org.openimaj.image.feature.local.aggregate.BlockSpatialAggregator;
import org.openimaj.io.IOUtils;
import org.openimaj.ml.annotation.ScoredAnnotation;
import org.openimaj.ml.annotation.linear.LiblinearAnnotator;
import org.openimaj.ml.clustering.ByteCentroidsResult;
import org.openimaj.ml.clustering.assignment.HardAssigner;
import org.openimaj.ml.clustering.kmeans.ByteKMeans;
import org.openimaj.util.pair.IntFloatPair;
import java.io.File;
import java.io.IOException;
import java.util.ArrayList;
import java.util.Date;
import java.util.List;
import java.util.Map;
/**
* Created by yschondorf on 5/29/2018.
*/
public class Chapter12Generic {
private static String IMAGES_PATH = "C:\\Development\\Video Analytics\\tpImages";
public static void main(String[] args) {
try {
LiblinearAnnotator<FImage, String> trainer = null;
VFSGroupDataset<FImage> allData = null;
allData = new VFSGroupDataset<FImage>(
IMAGES_PATH,
ImageUtilities.FIMAGE_READER);
GroupedDataset<String, ListDataset<FImage>, FImage> data =
GroupSampler.sample(allData, 1, false);
GroupedRandomSplitter<String, FImage> splits =
new GroupedRandomSplitter<String, FImage>(data, 15, 0, 15); // 15 training, 15 testing
DenseSIFT denseSIFT = new DenseSIFT(5, 7);
PyramidDenseSIFT<FImage> pyramidDenseSIFT = new PyramidDenseSIFT<FImage>(denseSIFT, 6f, 7);
GroupedDataset<String, ListDataset<FImage>, FImage> sample =
GroupedUniformRandomisedSampler.sample(splits.getTrainingDataset(), 15);
HardAssigner<byte[], float[], IntFloatPair> assigner = trainQuantiser(sample, pyramidDenseSIFT);
FeatureExtractor<DoubleFV, FImage> extractor = new PHOWExtractor(pyramidDenseSIFT, assigner);
//
// Now we’re ready to construct and train a classifier
//
trainer = new LiblinearAnnotator<FImage, String>(
extractor, LiblinearAnnotator.Mode.MULTICLASS, SolverType.L2R_L2LOSS_SVC, 1.0, 0.00001);
Date start = new Date();
System.out.println("Classifier training: start");
trainer.train(splits.getTrainingDataset());
System.out.println("Classifier training: end");
Date end = new Date();
long durationSec = (end.getTime() - start.getTime()) / 1000;
System.out.println("Classifier training duration: " + durationSec + " seconds");
final GroupedDataset<String, ListDataset<FImage>, FImage> testDataSet = splits.getTestDataset();
ClassificationEvaluator<CMResult<String>, String, FImage> eval =
new ClassificationEvaluator<CMResult<String>, String, FImage>(
trainer, testDataSet, new CMAnalyser<FImage, String>(CMAnalyser.Strategy.SINGLE));
start = new Date();
System.out.println("Classifier evaluation: start");
Map<FImage, ClassificationResult<String>> guesses = eval.evaluate();
System.out.println("Classifier evaluation - tp: end");
end = new Date();
durationSec = (end.getTime() - start.getTime()) / 1000;
System.out.println("Classifier evaluation duration: " + durationSec + " seconds");
CMResult<String> result = eval.analyse(guesses);
System.out.println("Result - tp: " + result);
} catch (IOException e) {
e.printStackTrace();
}
}
/**
* This method extracts the first 10000 dense SIFT features from the images in the dataset, and then clusters them
* into 300 separate classes. The method then returns a HardAssigner which can be used to assign SIFT features to
* identifiers
*
* @param pyramidDenseSIFT
* @return
*/
static HardAssigner<byte[], float[], IntFloatPair> trainQuantiser(
Dataset<FImage> sample,
// VFSGroupDataset<FImage> trainingImages,
PyramidDenseSIFT<FImage> pyramidDenseSIFT)
{
System.out.println("trainQuantiser: start");
Date start = new Date();
List<LocalFeatureList<ByteDSIFTKeypoint>> allKeys = new ArrayList<LocalFeatureList<ByteDSIFTKeypoint>>();
int i = 0;
int total = sample.numInstances();
// for (FImage image: sample) {
// ListDataset<FImage> images = trainingImages.get(key);
// total = images.size();
// break;
// }
for (FImage rec : sample) {
i++;
System.out.println(String.format("Analysing image %d out of %d", i, total));
FImage img = rec.getImage();
pyramidDenseSIFT.analyseImage(img);
allKeys.add(pyramidDenseSIFT.getByteKeypoints(0.005f));
}
final int numberOfDenseSiftFeaturesToExtract = 10000;
final int numberOfClassesInCluster = 300;
if (allKeys.size() > numberOfDenseSiftFeaturesToExtract)
allKeys = allKeys.subList(0, numberOfDenseSiftFeaturesToExtract);
ByteKMeans km = ByteKMeans.createKDTreeEnsemble(numberOfClassesInCluster);
DataSource<byte[]> dataSource = new LocalFeatureListDataSource<ByteDSIFTKeypoint, byte[]>(allKeys);
System.out.println(String.format(
"Clustering %d image features into %d classes...",
numberOfDenseSiftFeaturesToExtract, numberOfClassesInCluster));
ByteCentroidsResult result = km.cluster(dataSource);
Date end = new Date();
System.out.println("trainQuantiser: end");
System.out.println("trainQuantiser duration: " + (end.getTime() - start.getTime())/1000 + " seconds");
return result.defaultHardAssigner();
}
static class PHOWExtractor implements FeatureExtractor<DoubleFV, FImage> {
PyramidDenseSIFT<FImage> pdsift;
HardAssigner<byte[], float[], IntFloatPair> assigner;
public PHOWExtractor(PyramidDenseSIFT<FImage> pdsift, HardAssigner<byte[], float[], IntFloatPair> assigner)
{
this.pdsift = pdsift;
this.assigner = assigner;
}
public DoubleFV extractFeature(FImage object) {
FImage image = object.getImage();
pdsift.analyseImage(image);
BagOfVisualWords<byte[]> bovw = new BagOfVisualWords<byte[]>(assigner);
BlockSpatialAggregator<byte[], SparseIntFV> spatial = new BlockSpatialAggregator<byte[], SparseIntFV>(
bovw, 2, 2);
return spatial.aggregate(pdsift.getByteKeypoints(0.015f), image.getBounds()).normaliseFV();
}
}
}
package com.mycompany.video.analytics;
import de.bwaldvogel.liblinear.SolverType;
import org.apache.commons.vfs2.FileSystemException;
import org.openimaj.data.DataSource;
import org.openimaj.data.dataset.Dataset;
import org.openimaj.data.dataset.GroupedDataset;
import org.openimaj.data.dataset.ListDataset;
import org.openimaj.data.dataset.VFSGroupDataset;
import org.openimaj.experiment.dataset.sampling.GroupSampler;
import org.openimaj.experiment.dataset.sampling.GroupedUniformRandomisedSampler;
import org.openimaj.experiment.dataset.split.GroupedRandomSplitter;
import org.openimaj.experiment.evaluation.classification.ClassificationEvaluator;
import org.openimaj.experiment.evaluation.classification.ClassificationResult;
import org.openimaj.experiment.evaluation.classification.analysers.confusionmatrix.CMAnalyser;
import org.openimaj.experiment.evaluation.classification.analysers.confusionmatrix.CMResult;
import org.openimaj.feature.DoubleFV;
import org.openimaj.feature.FeatureExtractor;
import org.openimaj.feature.SparseIntFV;
import org.openimaj.feature.local.data.LocalFeatureListDataSource;
import org.openimaj.feature.local.list.LocalFeatureList;
import org.openimaj.image.FImage;
import org.openimaj.image.ImageUtilities;
import org.openimaj.image.feature.dense.gradient.dsift.ByteDSIFTKeypoint;
import org.openimaj.image.feature.dense.gradient.dsift.DenseSIFT;
import org.openimaj.image.feature.dense.gradient.dsift.PyramidDenseSIFT;
import org.openimaj.image.feature.local.aggregate.BagOfVisualWords;
import org.openimaj.image.feature.local.aggregate.BlockSpatialAggregator;
import org.openimaj.io.IOUtils;
import org.openimaj.ml.annotation.ScoredAnnotation;
import org.openimaj.ml.annotation.linear.LiblinearAnnotator;
import org.openimaj.ml.clustering.ByteCentroidsResult;
import org.openimaj.ml.clustering.assignment.HardAssigner;
import org.openimaj.ml.clustering.kmeans.ByteKMeans;
import org.openimaj.util.pair.IntFloatPair;
import java.io.File;
import java.io.IOException;
import java.util.ArrayList;
import java.util.Date;
import java.util.List;
import java.util.Map;
public class Chapter12Generic {
private static String IMAGES_PATH = "C:\\Development\\Video Analytics\\tpImages";
private static String TEST_IMAGES_PATH = "C:\\Development\\Video Analytics\\testImages";
private static String TRAINER_DATA_FILE_PATH = "C:\\Development\\Video Analytics\\out\\trainer.dat";
public static void main(String[] args) throws Exception {
LiblinearAnnotator<FImage, String> trainer = null;
File inputDataFile = new File(TRAINER_DATA_FILE_PATH);
if (inputDataFile.isFile()) {
trainer = IOUtils.readFromFile(inputDataFile);
} else {
VFSGroupDataset<FImage> allData = null;
allData = new VFSGroupDataset<FImage>(
IMAGES_PATH,
ImageUtilities.FIMAGE_READER);
GroupedDataset<String, ListDataset<FImage>, FImage> data =
GroupSampler.sample(allData, 1, false);
GroupedRandomSplitter<String, FImage> splits =
new GroupedRandomSplitter<String, FImage>(data, 15, 0, 15); // 15 training, 15 testing
DenseSIFT denseSIFT = new DenseSIFT(5, 7);
PyramidDenseSIFT<FImage> pyramidDenseSIFT = new PyramidDenseSIFT<FImage>(denseSIFT, 6f, 7);
GroupedDataset<String, ListDataset<FImage>, FImage> sample =
GroupedUniformRandomisedSampler.sample(splits.getTrainingDataset(), 15);
HardAssigner<byte[], float[], IntFloatPair> assigner = trainQuantiser(sample, pyramidDenseSIFT);
FeatureExtractor<DoubleFV, FImage> extractor = new PHOWExtractor(pyramidDenseSIFT, assigner);
//
// Now we’re ready to construct and train a classifier
//
trainer = new LiblinearAnnotator<FImage, String>(
extractor, LiblinearAnnotator.Mode.MULTICLASS, SolverType.L2R_L2LOSS_SVC, 1.0, 0.00001);
Date start = new Date();
System.out.println("Classifier training: start");
trainer.train(splits.getTrainingDataset());
IOUtils.writeToFile(trainer, inputDataFile);
System.out.println("Classifier training: end");
Date end = new Date();
long durationSec = (end.getTime() - start.getTime()) / 1000;
System.out.println("Classifier training duration: " + durationSec + " seconds");
}
// final GroupedDataset<String, ListDataset<FImage>, FImage> testDataSet = splits.getTestDataset();
VFSGroupDataset<FImage> testDataSet = new VFSGroupDataset<FImage>(
TEST_IMAGES_PATH,
ImageUtilities.FIMAGE_READER);
ClassificationEvaluator<CMResult<String>, String, FImage> eval =
new ClassificationEvaluator<CMResult<String>, String, FImage>(
trainer, testDataSet, new CMAnalyser<FImage, String>(CMAnalyser.Strategy.SINGLE));
Date start = new Date();
System.out.println("Classifier evaluation: start");
Map<FImage, ClassificationResult<String>> guesses = eval.evaluate();
System.out.println("Classifier evaluation - tp: end");
Date end = new Date();
long durationSec = (end.getTime() - start.getTime()) / 1000;
System.out.println("Classifier evaluation duration: " + durationSec + " seconds");
CMResult<String> result = eval.analyse(guesses);
System.out.println("Result - tp: " + result);
}
/**
* This method extracts the first 10000 dense SIFT features from the images in the dataset, and then clusters them
* into 300 separate classes. The method then returns a HardAssigner which can be used to assign SIFT features to
* identifiers
*
* @param pyramidDenseSIFT
* @return
*/
static HardAssigner<byte[], float[], IntFloatPair> trainQuantiser(
Dataset<FImage> sample,
// VFSGroupDataset<FImage> trainingImages,
PyramidDenseSIFT<FImage> pyramidDenseSIFT)
{
System.out.println("trainQuantiser: start");
Date start = new Date();
List<LocalFeatureList<ByteDSIFTKeypoint>> allKeys = new ArrayList<LocalFeatureList<ByteDSIFTKeypoint>>();
int i = 0;
int total = sample.numInstances();
// for (FImage image: sample) {
// ListDataset<FImage> images = trainingImages.get(key);
// total = images.size();
// break;
// }
for (FImage rec : sample) {
i++;
System.out.println(String.format("Analysing image %d out of %d", i, total));
FImage img = rec.getImage();
pyramidDenseSIFT.analyseImage(img);
allKeys.add(pyramidDenseSIFT.getByteKeypoints(0.005f));
}
final int numberOfDenseSiftFeaturesToExtract = 10000;
final int numberOfClassesInCluster = 300;
if (allKeys.size() > numberOfDenseSiftFeaturesToExtract)
allKeys = allKeys.subList(0, numberOfDenseSiftFeaturesToExtract);
ByteKMeans km = ByteKMeans.createKDTreeEnsemble(numberOfClassesInCluster);
DataSource<byte[]> dataSource = new LocalFeatureListDataSource<ByteDSIFTKeypoint, byte[]>(allKeys);
System.out.println(String.format(
"Clustering %d image features into %d classes...",
numberOfDenseSiftFeaturesToExtract, numberOfClassesInCluster));
ByteCentroidsResult result = km.cluster(dataSource);
Date end = new Date();
System.out.println("trainQuantiser: end");
System.out.println("trainQuantiser duration: " + (end.getTime() - start.getTime())/1000 + " seconds");
return result.defaultHardAssigner();
}
static class PHOWExtractor implements FeatureExtractor<DoubleFV, FImage> {
PyramidDenseSIFT<FImage> pdsift;
HardAssigner<byte[], float[], IntFloatPair> assigner;
public PHOWExtractor(PyramidDenseSIFT<FImage> pdsift, HardAssigner<byte[], float[], IntFloatPair> assigner)
{
this.pdsift = pdsift;
this.assigner = assigner;
}
public DoubleFV extractFeature(FImage object) {
FImage image = object.getImage();
pdsift.analyseImage(image);
BagOfVisualWords<byte[]> bovw = new BagOfVisualWords<byte[]>(assigner);
BlockSpatialAggregator<byte[], SparseIntFV> spatial = new BlockSpatialAggregator<byte[], SparseIntFV>(
bovw, 2, 2);
return spatial.aggregate(pdsift.getByteKeypoints(0.015f), image.getBounds()).normaliseFV();
}
}
}
CMResult<String> result = eval.analyse(guesses);
public class Chapter12Generic_v3 {
// contains an accordion folder with images from caltech101
private static String TRAINING_IMAGES_PATH = "C:\\Development\\Video Analytics\\images";
// contains 1 airplane image from caltech101
private static String TEST_IMAGE = "C:\\Development\\Video Analytics\\testImages\\falseImages\\image_0001.jpg";
private static String TRAINER_DATA_FILE_PATH = "C:\\Development\\Video Analytics\\out\\trainer.dat";
public static void main(String[] args) throws Exception {
LiblinearAnnotator<FImage, String> trainer = null;
File inputDataFile = new File(TRAINER_DATA_FILE_PATH);
if (inputDataFile.isFile()) {
trainer = IOUtils.readFromFile(inputDataFile);
} else {
VFSGroupDataset<FImage> allData = null;
allData = new VFSGroupDataset<FImage>(
TRAINING_IMAGES_PATH,
ImageUtilities.FIMAGE_READER);
GroupedDataset<String, ListDataset<FImage>, FImage> data =
GroupSampler.sample(allData, 1, false);
GroupedRandomSplitter<String, FImage> splits =
new GroupedRandomSplitter<String, FImage>(data, 15, 0, 15); // 15 training, 15 testing
DenseSIFT denseSIFT = new DenseSIFT(5, 7);
PyramidDenseSIFT<FImage> pyramidDenseSIFT = new PyramidDenseSIFT<FImage>(denseSIFT, 6f, 7);
GroupedDataset<String, ListDataset<FImage>, FImage> sample =
GroupedUniformRandomisedSampler.sample(splits.getTrainingDataset(), 15);
HardAssigner<byte[], float[], IntFloatPair> assigner = trainQuantiser(sample, pyramidDenseSIFT);
FeatureExtractor<DoubleFV, FImage> extractor = new PHOWExtractor(pyramidDenseSIFT, assigner);
//
// Now we’re ready to construct and train a classifier
//
trainer = new LiblinearAnnotator<FImage, String>(
extractor, LiblinearAnnotator.Mode.MULTICLASS, SolverType.L2R_L2LOSS_SVC, 1.0, 0.00001);
Date start = new Date();
System.out.println("Classifier training: start");
trainer.train(splits.getTrainingDataset());
IOUtils.writeToFile(trainer, new File(TRAINER_DATA_FILE_PATH));
System.out.println("Classifier training: end");
Date end = new Date();
long durationSec = (end.getTime() - start.getTime()) / 1000;
System.out.println("Classifier training duration: " + durationSec + " seconds");
}
FImage query = ImageUtilities.readF(new File(TEST_IMAGE));
final List<ScoredAnnotation<String>> scoredAnnotations = trainer.annotate(query);
final ClassificationResult<String> classificationResult = trainer.classify(query);
System.out.println("scoredAnnotations: " + scoredAnnotations);
System.out.println("classificationResult: " + classificationResult);
}
/**
* This method extracts the first 10000 dense SIFT features from the images in the dataset, and then clusters them
* into 300 separate classes. The method then returns a HardAssigner which can be used to assign SIFT features to
* identifiers
*
* @param pyramidDenseSIFT
* @return
*/
static HardAssigner<byte[], float[], IntFloatPair> trainQuantiser(
Dataset<FImage> sample,
PyramidDenseSIFT<FImage> pyramidDenseSIFT)
{
System.out.println("trainQuantiser: start");
Date start = new Date();
List<LocalFeatureList<ByteDSIFTKeypoint>> allKeys = new ArrayList<LocalFeatureList<ByteDSIFTKeypoint>>();
int i = 0;
int total = sample.numInstances();
for (FImage rec : sample) {
i++;
System.out.println(String.format("Analysing image %d out of %d", i, total));
FImage img = rec.getImage();
pyramidDenseSIFT.analyseImage(img);
allKeys.add(pyramidDenseSIFT.getByteKeypoints(0.005f));
}
final int numberOfDenseSiftFeaturesToExtract = 10000;
final int numberOfClassesInCluster = 300;
if (allKeys.size() > numberOfDenseSiftFeaturesToExtract)
allKeys = allKeys.subList(0, numberOfDenseSiftFeaturesToExtract);
ByteKMeans km = ByteKMeans.createKDTreeEnsemble(numberOfClassesInCluster);
DataSource<byte[]> dataSource = new LocalFeatureListDataSource<ByteDSIFTKeypoint, byte[]>(allKeys);
System.out.println(String.format(
"Clustering %d image features into %d classes...",
numberOfDenseSiftFeaturesToExtract, numberOfClassesInCluster));
ByteCentroidsResult result = km.cluster(dataSource);
Date end = new Date();
System.out.println("trainQuantiser: end");
System.out.println("trainQuantiser duration: " + (end.getTime() - start.getTime())/1000 + " seconds");
return result.defaultHardAssigner();
}
static class PHOWExtractor implements FeatureExtractor<DoubleFV, FImage> {
PyramidDenseSIFT<FImage> pdsift;
HardAssigner<byte[], float[], IntFloatPair> assigner;
public PHOWExtractor(PyramidDenseSIFT<FImage> pdsift, HardAssigner<byte[], float[], IntFloatPair> assigner)
{
this.pdsift = pdsift;
this.assigner = assigner;
}
public DoubleFV extractFeature(FImage object) {
FImage image = object.getImage();
pdsift.analyseImage(image);
BagOfVisualWords<byte[]> bovw = new BagOfVisualWords<byte[]>(assigner);
BlockSpatialAggregator<byte[], SparseIntFV> spatial = new BlockSpatialAggregator<byte[], SparseIntFV>(
bovw, 2, 2);
return spatial.aggregate(pdsift.getByteKeypoints(0.015f), image.getBounds()).normaliseFV();
}
}
}
如果你只是想用你训练过的模型对事物进行分类，那么忽略所有的ClassificationEvaluator东西——这只是为了计算的准确性，等等
查看培训师对象的类型。当您的培训师实例在FImage上键入并字符串时，其注释和分类方法都将接受您提供的FImage作为输入，并以稍微不同的形式提供分类结果作为输出；你必须决定哪一个最适合你的需要。你有一个测试数据集就在那里-我不知道那是测试图像所在的位置-所以你必须将图像粘贴在那里-你必须通过documentation@gpasch我根据你的评论更新了这个问题。除了教程之外，我没有找到任何重要的文档。我不能使用testDataset，因为我想将训练部分与检测部分分开，而新代码会抛出一个NullPointerException。我正在关注这个ClassificationEvaluator eval=new ClassificationEvaluator trainer，testDataset，new CmAnalyserCmanAnalyser.Strategy.SINGLE；-您必须查看文档中的ClassificationEvaluator不要期望教程中的所有内容-因此，是的，您需要进行培训并存储结果。然后用你想要的任何东西进行测试——如果必须的话，替换整个测试集。我能找到的唯一文档就是API引用。eval有两种方法可以使用：evaluate和analyseMap predicted，因此我应该按顺序执行它们：Map guesses=eval.evaluate；CMResult结果=评估分析结果；但是第二个调用抛出了一个异常，正如我已经提到的……我根据您的答案@Jon添加了代码的版本3。主要的变化是实例化单个FImage，并调用trainer.annotate。现在的问题是，我放了一个飞机测试图像，但它被归类为手风琴。TRAINING_IMAGES_PATH包含一个包含caltech101手风琴图像的手风琴文件夹。TEST_图像包含一张来自caltech101的飞机图像。什么仍然不正确？你不能只在一节课上训练它-在训练过程中，你至少需要有一节带示例图片的负面课你是对的！我为飞机添加了另一个文件夹，更改了行GroupedData=GroupSampler.sampleallData，1，false；to GroupedDataset data=GroupSampler.sampleallData，2，false；毫无疑问，飞机图像的分类是正确的。
CMResult<String> result = eval.analyse(guesses);
public class Chapter12Generic_v3 {
// contains an accordion folder with images from caltech101
private static String TRAINING_IMAGES_PATH = "C:\\Development\\Video Analytics\\images";
// contains 1 airplane image from caltech101
private static String TEST_IMAGE = "C:\\Development\\Video Analytics\\testImages\\falseImages\\image_0001.jpg";
private static String TRAINER_DATA_FILE_PATH = "C:\\Development\\Video Analytics\\out\\trainer.dat";
public static void main(String[] args) throws Exception {
LiblinearAnnotator<FImage, String> trainer = null;
File inputDataFile = new File(TRAINER_DATA_FILE_PATH);
if (inputDataFile.isFile()) {
trainer = IOUtils.readFromFile(inputDataFile);
} else {
VFSGroupDataset<FImage> allData = null;
allData = new VFSGroupDataset<FImage>(
TRAINING_IMAGES_PATH,
ImageUtilities.FIMAGE_READER);
GroupedDataset<String, ListDataset<FImage>, FImage> data =
GroupSampler.sample(allData, 1, false);
GroupedRandomSplitter<String, FImage> splits =
new GroupedRandomSplitter<String, FImage>(data, 15, 0, 15); // 15 training, 15 testing
DenseSIFT denseSIFT = new DenseSIFT(5, 7);
PyramidDenseSIFT<FImage> pyramidDenseSIFT = new PyramidDenseSIFT<FImage>(denseSIFT, 6f, 7);
GroupedDataset<String, ListDataset<FImage>, FImage> sample =
GroupedUniformRandomisedSampler.sample(splits.getTrainingDataset(), 15);
HardAssigner<byte[], float[], IntFloatPair> assigner = trainQuantiser(sample, pyramidDenseSIFT);
FeatureExtractor<DoubleFV, FImage> extractor = new PHOWExtractor(pyramidDenseSIFT, assigner);
//
// Now we’re ready to construct and train a classifier
//
trainer = new LiblinearAnnotator<FImage, String>(
extractor, LiblinearAnnotator.Mode.MULTICLASS, SolverType.L2R_L2LOSS_SVC, 1.0, 0.00001);
Date start = new Date();
System.out.println("Classifier training: start");
trainer.train(splits.getTrainingDataset());
IOUtils.writeToFile(trainer, new File(TRAINER_DATA_FILE_PATH));
System.out.println("Classifier training: end");
Date end = new Date();
long durationSec = (end.getTime() - start.getTime()) / 1000;
System.out.println("Classifier training duration: " + durationSec + " seconds");
}
FImage query = ImageUtilities.readF(new File(TEST_IMAGE));
final List<ScoredAnnotation<String>> scoredAnnotations = trainer.annotate(query);
final ClassificationResult<String> classificationResult = trainer.classify(query);
System.out.println("scoredAnnotations: " + scoredAnnotations);
System.out.println("classificationResult: " + classificationResult);
}
/**
* This method extracts the first 10000 dense SIFT features from the images in the dataset, and then clusters them
* into 300 separate classes. The method then returns a HardAssigner which can be used to assign SIFT features to
* identifiers
*
* @param pyramidDenseSIFT
* @return
*/
static HardAssigner<byte[], float[], IntFloatPair> trainQuantiser(
Dataset<FImage> sample,
PyramidDenseSIFT<FImage> pyramidDenseSIFT)
{
System.out.println("trainQuantiser: start");
Date start = new Date();
List<LocalFeatureList<ByteDSIFTKeypoint>> allKeys = new ArrayList<LocalFeatureList<ByteDSIFTKeypoint>>();
int i = 0;
int total = sample.numInstances();
for (FImage rec : sample) {
i++;
System.out.println(String.format("Analysing image %d out of %d", i, total));
FImage img = rec.getImage();
pyramidDenseSIFT.analyseImage(img);
allKeys.add(pyramidDenseSIFT.getByteKeypoints(0.005f));
}
final int numberOfDenseSiftFeaturesToExtract = 10000;
final int numberOfClassesInCluster = 300;
if (allKeys.size() > numberOfDenseSiftFeaturesToExtract)
allKeys = allKeys.subList(0, numberOfDenseSiftFeaturesToExtract);
ByteKMeans km = ByteKMeans.createKDTreeEnsemble(numberOfClassesInCluster);
DataSource<byte[]> dataSource = new LocalFeatureListDataSource<ByteDSIFTKeypoint, byte[]>(allKeys);
System.out.println(String.format(
"Clustering %d image features into %d classes...",
numberOfDenseSiftFeaturesToExtract, numberOfClassesInCluster));
ByteCentroidsResult result = km.cluster(dataSource);
Date end = new Date();
System.out.println("trainQuantiser: end");
System.out.println("trainQuantiser duration: " + (end.getTime() - start.getTime())/1000 + " seconds");
return result.defaultHardAssigner();
}
static class PHOWExtractor implements FeatureExtractor<DoubleFV, FImage> {
PyramidDenseSIFT<FImage> pdsift;
HardAssigner<byte[], float[], IntFloatPair> assigner;
public PHOWExtractor(PyramidDenseSIFT<FImage> pdsift, HardAssigner<byte[], float[], IntFloatPair> assigner)
{
this.pdsift = pdsift;
this.assigner = assigner;
}
public DoubleFV extractFeature(FImage object) {
FImage image = object.getImage();
pdsift.analyseImage(image);
BagOfVisualWords<byte[]> bovw = new BagOfVisualWords<byte[]>(assigner);
BlockSpatialAggregator<byte[], SparseIntFV> spatial = new BlockSpatialAggregator<byte[], SparseIntFV>(
bovw, 2, 2);
return spatial.aggregate(pdsift.getByteKeypoints(0.015f), image.getBounds()).normaliseFV();
}
}
}