def getLatentSemantic(k, decompType, dataMatrix, modelType, label,
imageDirName, imagePaths):
folderName = "{}_{}_{}_{}_{}".format(imageDirName, modelType.name,
decompType.name, k, label)
lsPath = getLatentSemanticPath(os.path.basename(imageDirName), modelType,
decompType, k, label)
latent_semantic = latentSemanticsHelper.getSemanticsFromFolder(lsPath)
if latent_semantic is None:
if decompType == reduction.ReductionType.SVD:
u, v = SVD(dataMatrix, k).getDecomposition()
latent_semantic = u, v
elif decompType == reduction.ReductionType.PCA:
latent_semantic = PCA(dataMatrix, k).getDecomposition()
elif decompType == reduction.ReductionType.NMF:
latent_semantic = NMF(dataMatrix, k).getDecomposition()
elif decompType == reduction.ReductionType.LDA:
latent_semantic = LDA(dataMatrix, k).getDecomposition()
else:
print("Check later")
return None
print("Image path example ", imagePaths[0])
latentSemanticsHelper.saveSemantics(os.path.basename(imageDirName),
modelType,
label,
decompType,
k,
latent_semantic[0],
latent_semantic[1],
imagePaths=imagePaths)
return latent_semantic
def initTask5_2(folderPath, imagePath):
classificatonMeta = {
"Dorsal": "palmar",
"left": "Right-Handed",
"With-Accessories": "Without-Accessories",
"male": "female",
"palmar": "Dorsal",
"Right-Handed": "left",
"Without-Accessories": "With-Accessories",
"female": "male"
}
_, modelType, dimRedType, k, label = getParams(folderPath)
# print(folderPath)
u, vt, imagePaths = getSemanticsFromFolder(folderPath)
u = preprocessing.scale(u)
uMean = np.mean(u, axis=0)
maxdis, mindis = -10000000, 10000000
for item in u:
d = np.linalg.norm(item - uMean)
mindis = min(mindis, d)
maxdis = max(maxdis, d)
print(mindis, maxdis, label)
queryImage = getQueryImageRep(vt, imagePath, modelType)
queryImageNormalised = preprocessing.scale(queryImage)
qdis = np.linalg.norm(queryImageNormalised - uMean)
print(qdis)
def task4(foldername, folderPath, imagePath, m):
print(" EXECUTING TASK 4 ")
# print(folderPath)
# print("Image Path:" + imagePath)
U, V, imagePaths = getSemanticsFromFolder(folderPath)
dir, modelType, dimRidTechnique, K, label = getParams(foldername)
query_image_features = getQueryImageRep(V, imagePath, modelType)
list = comparisonHelper.getMSimilarImages(U, query_image_features, m,
imagePaths)
plotFigures(list, 3)
def initTask5(folderPath, imagePath):
classificatonMeta = {
"Dorsal": "Palmer",
"Left-Handed": "Right-Handed",
"With-Accessories": "Without-Accessories",
"Male": "Female",
"Palmer": "Dorsal",
"Right-Handed": "Left-Handed",
"Without-Accessories": "With-Accessories",
"Female": "Male"
}
labelInfo = {
"1": "Left-Handed",
"2": "Right-Handed",
"3": "Dorsal",
"4": "Palmer",
"5": "With-Accessories",
"6": "Without-Accessories",
"7": "Male",
"8": "Female"
}
_, modelType, dimRedType, k, label = getParams(folderPath)
u, vt, imagePaths = getSemanticsFromFolder(folderPath)
uNomalised = preprocessing.scale(u)
oc_svm_clf = svm.OneClassSVM(gamma=0.01, kernel='rbf', nu=0.1)
oc_svm_clf.fit(uNomalised)
queryImage = getQueryImageRep(vt, imagePath, modelType)
queryImageNormalised = preprocessing.scale(queryImage)
queryPrediction = oc_svm_clf.predict(
queryImageNormalised.reshape((1, queryImageNormalised.shape[0])))
if len(queryPrediction) < 1:
raise ValueError("Query prediction not available")
if queryPrediction[0] == 1:
print("Predicted the image as '{}'".format(labelInfo[label]))
else:
print("Predicted the image as '{}'".format(
classificatonMeta[labelInfo[label]]))