def plot_curve():
io.log('Loading cross-validation values...')
cv_values = io.load_object('intersection_svm_CNNfeatures', ignore=True)
io.log('Loading best parameters...')
best_params = io.load_object('best_params_intersection_svm_CNNfeatures1',
ignore=True)
io.log('Plotting...')
colors = itertools.cycle([
'blue', 'green', 'red', 'cyan', 'magenta', 'yellow', 'darkolivegreen',
'darkviolet', 'black'
])
# Subplot parameters
plt.figure(facecolor='white')
num_subplots = len(codebook_size)
num_columns = 1
num_rows = np.ceil(num_subplots / num_columns)
# All subplots
for ind, k in enumerate(codebook_size):
# Search dictionary
val = cv_values[(k)]
results = val['cv_results']
feature_time = val['feature_time']
pca_time = val['pca_time']
gmm_time = val['gmm_time']
fisher_time = val['fisher_time']
scaler_time = val['scaler_time']
crossvalidation_time = val['crossvalidation_time']
total_time = val['total_time']
# Plot
x = results['param_C']
y = results['mean_test_score']
e = results['std_test_score']
sorted_indices = x.argsort()
x_sorted = np.asarray(x[sorted_indices], dtype=np.float64)
y_sorted = np.asarray(y[sorted_indices], dtype=np.float64)
e_sorted = np.asarray(e[sorted_indices], dtype=np.float64)
color = colors.next()
ax = plt.subplot(num_rows, num_columns, ind + 1)
ax.set_xscale("log")
ax.set_ylim((0.7, 0.9))
ax.errorbar(x_sorted,
y_sorted,
e_sorted,
linestyle='--',
lw=2,
marker='x',
color=color)
ax.set_title('{} Gaussians in GMM'.format(k))
ax.set_xlabel('C')
ax.set_ylabel('Accuracy')
# Print information
io.log('CODEBOOK {} '.format(k))
io.log('-------------')
io.log('Mean accuracy: {}'.format(y.max()))
io.log('Std accuracy: {}'.format(e[np.argmax(y)]))
io.log('C: {}'.format(x[np.argmax(y)]))
io.log()
io.log('Timing')
io.log('\tSIFT time: {:.2f} s'.format(feature_time))
io.log('\tPCA time: {:.2f} s'.format(pca_time))
io.log('\tGMM time: {:.2f} s'.format(gmm_time))
io.log('\tFisher time: {:.2f} s'.format(fisher_time))
io.log('\tScaler time: {:.2f} s'.format(scaler_time))
io.log('\tCV time: {:.2f} s'.format(crossvalidation_time))
io.log('\t_________________________')
io.log('\tTOTAL TIME: {:.2f} s'.format(total_time))
io.log()
plt.tight_layout()
plt.show()
plt.close()
def train():
best_accuracy = 0
best_params = {}
cv_results = {}
base_model = VGG16(weights='imagenet')
# crop the model up to a certain layer
model = Model(input=base_model.input,
output=base_model.get_layer('block5_conv2').output)
# Read the training set
train_images_filenames = cPickle.load(
open('./dataset/train_images_filenames.dat', 'r'))
test_images_filenames = cPickle.load(
open('./dataset/test_images_filenames.dat', 'r'))
train_labels = cPickle.load(open('./dataset/train_labels.dat', 'r'))
test_labels = cPickle.load(open('./dataset/test_labels.dat', 'r'))
io.log('\nLoaded {} train images.'.format(len(train_images_filenames)))
io.log('\nLoaded {} test images.'.format(len(test_images_filenames)))
# read and process training images
print 'Getting features from training images'
start_feature = time.time()
Train_descriptors = []
Train_label_per_descriptor = []
for i in range(len(train_images_filenames)):
img = image.load_img(train_images_filenames[i], target_size=(224, 224))
x = image.img_to_array(img)
x = np.expand_dims(x, axis=0)
x = preprocess_input(x)
# get the features from images
features = model.predict(x)
features = features[0, :, :, :]
descriptor = features.reshape(features.shape[0] * features.shape[1],
features.shape[2])
Train_descriptors.append(descriptor)
Train_label_per_descriptor.append(train_labels[i])
# Put all descriptors in a numpy array to compute PCA and GMM
size_descriptors = Train_descriptors[0].shape[1]
Desc = np.zeros(
(np.sum([len(p) for p in Train_descriptors]), size_descriptors),
dtype=np.uint8)
startingpoint = 0
for i in range(len(Train_descriptors)):
Desc[startingpoint:startingpoint +
len(Train_descriptors[i])] = Train_descriptors[i]
startingpoint += len(Train_descriptors[i])
feature_time = time.time() - start_feature
io.log('Elapsed time: {:.2f} s'.format(feature_time))
for dim_red in pca_reduction:
io.log('Applying PCA ... ')
start_pca = time.time()
reduction = np.int(dim_red * Desc.shape[1])
pca = decomposition.PCA(n_components=reduction)
pca.fit(Desc)
Desc_pca = np.float32(pca.transform(Desc))
pca_time = time.time() - start_pca
io.log('Elapsed time: {:.2f} s'.format(pca_time))
for k in codebook_size:
io.log('Creating GMM model (k = {})'.format(k))
start_gmm = time.time()
gmm = ynumpy.gmm_learn(np.float32(Desc_pca), k)
io.save_object(gmm, 'gmm_NN_pca_{}_k_{}'.format(reduction, k))
gmm_time = time.time() - start_gmm
io.log('Elapsed time: {:.2f} s'.format(gmm_time))
io.log('Getting Fisher vectors from training set...')
start_fisher = time.time()
fisher = np.zeros((len(Train_descriptors), k * reduction * 2),
dtype=np.float32)
for i in xrange(len(Train_descriptors)):
descriptor = Train_descriptors[i]
descriptor = np.float32(pca.transform(descriptor))
fisher[i, :] = ynumpy.fisher(gmm,
descriptor,
include=['mu', 'sigma'])
# L2 normalization - reshape to avoid deprecation warning, checked that the result is the same
fisher[i, :] = preprocessing.normalize(fisher[i, :].reshape(
1, -1),
norm='l2')
fisher_time = time.time() - start_fisher
io.log('Elapsed time: {:.2f} s'.format(fisher_time))
io.log('Scaling features...')
start_scaler = time.time()
stdSlr = StandardScaler().fit(fisher)
D_scaled = stdSlr.transform(fisher)
scaler_time = time.time() - start_scaler
io.log('Elapsed time: {:.2f} s'.format(scaler_time))
io.log('Optimizing SVM hyperparameters...')
start_crossvalidation = time.time()
svm = SVC(kernel='precomputed')
random_search = RandomizedSearchCV(svm,
params_distribution,
n_iter=n_iter,
scoring='accuracy',
refit=False,
cv=3,
verbose=1)
# Precompute Gram matrix
gram = kernels.intersection_kernel(D_scaled, D_scaled)
random_search.fit(gram, train_labels)
crossvalidation_time = time.time() - start_crossvalidation
io.log('Elapsed time: {:.2f} s'.format(crossvalidation_time))
# Convert MaskedArrays to ndarrays to avoid unpickling bugs
results = random_search.cv_results_
results['param_C'] = results['param_C'].data
# Appending all parameter-scores combinations
cv_results.update({
(k): {
'cv_results':
results,
'feature_time':
feature_time,
'pca_time':
pca_time,
'gmm_time':
gmm_time,
'fisher_time':
fisher_time,
'scaler_time':
scaler_time,
'crossvalidation_time':
crossvalidation_time,
'total_time':
feature_time + pca_time + gmm_time + fisher_time +
scaler_time + crossvalidation_time
}
})
io.save_object(cv_results,
'intersection_svm_CNNfeatures',
ignore=True)
# Obtaining the parameters which yielded the best accuracy
if random_search.best_score_ > best_accuracy:
best_accuracy = random_search.best_score_
best_params = random_search.best_params_
best_params.update({'k': k, 'pca': dim_red})
io.log('-------------------------------\n')
io.log('\nSaving best parameters...')
io.save_object(best_params,
'best_params_intersection_svm_CNNfeatures',
ignore=True)
best_params_file = os.path.abspath(
'./ignore/best_params_intersection_svm_CNNfeatures.pickle')
io.log('Saved at {}'.format(best_params_file))
io.log('\nSaving all cross-validation values...')
io.save_object(cv_results, 'intersection_svm_CNNfeatures', ignore=True)
cv_results_file = os.path.abspath(
'./ignore/intersection_svm_CNNfeatures.pickle')
io.log('Saved at {}'.format(cv_results_file))
io.log('\nBEST PARAMS')
io.log('k={}, dim_red={}, C={} --> accuracy: {:.3f}'.format(
best_params['k'], best_params['pca'], best_params['C'], best_accuracy))
def train():
best_accuracy = 0
best_params = {}
cv_results = {}
base_model = VGG16(weights='imagenet')
# crop the model up to a certain layer
model = Model(input=base_model.input,
output=base_model.get_layer('fc2').output)
# Read the training set
train_images_filenames = cPickle.load(
open('./dataset/train_images_filenames.dat', 'r'))
test_images_filenames = cPickle.load(
open('./dataset/test_images_filenames.dat', 'r'))
train_labels = cPickle.load(open('./dataset/train_labels.dat', 'r'))
test_labels = cPickle.load(open('./dataset/test_labels.dat', 'r'))
io.log('\nLoaded {} train images.'.format(len(train_images_filenames)))
io.log('\nLoaded {} test images.'.format(len(test_images_filenames)))
# read and process training images
print 'Getting features from training images'
start_feature = time.time()
first = 1
for i in range(len(train_images_filenames)):
img = image.load_img(train_images_filenames[i], target_size=(224, 224))
x = image.img_to_array(img)
x = np.expand_dims(x, axis=0)
x = preprocess_input(x)
# get the features from images
features = model.predict(x)
features = features[0, :]
if first == 1:
Desc = features
first = 0
else:
Desc = np.vstack((Desc, features))
feature_time = time.time() - start_feature
io.log('Elapsed time: {:.2f} s'.format(feature_time))
io.log('Scaling features...')
start_scaler = time.time()
stdSlr = StandardScaler().fit(Desc)
D_scaled = stdSlr.transform(Desc)
scaler_time = time.time() - start_scaler
io.log('Elapsed time: {:.2f} s'.format(scaler_time))
io.log('Optimizing SVM hyperparameters...')
start_crossvalidation = time.time()
svm = SVC(kernel='precomputed', probability=True)
random_search = RandomizedSearchCV(svm,
params_distribution,
n_iter=n_iter,
scoring='accuracy',
refit=False,
cv=3,
verbose=1)
# Precompute Gram matrix
gram = kernels.intersection_kernel(D_scaled, D_scaled)
random_search.fit(gram, train_labels)
crossvalidation_time = time.time() - start_crossvalidation
io.log('Elapsed time: {:.2f} s'.format(crossvalidation_time))
# Convert MaskedArrays to ndarrays to avoid unpickling bugs
results = random_search.cv_results_
results['param_C'] = results['param_C'].data
# Appending all parameter-scores combinations
cv_results.update({
'cv_results':
results,
'feature_time':
feature_time,
'scaler_time':
scaler_time,
'crossvalidation_time':
crossvalidation_time,
'total_time':
feature_time + scaler_time + crossvalidation_time
})
io.save_object(cv_results, 'intersection_svm_CNNfeatures', ignore=True)
print('Best accuracy ' + str(random_search.best_score_))
# Obtaining the parameters which yielded the best accuracy
if random_search.best_score_ > best_accuracy:
best_accuracy = random_search.best_score_
best_params = random_search.best_params_
io.log('-------------------------------\n')
io.log('\nSaving best parameters...')
io.save_object(best_params,
'best_params_intersection_svm_CNNfeatures',
ignore=True)
best_params_file = os.path.abspath(
'./ignore/best_params_intersection_svm_CNNfeatures.pickle')
io.log('Saved at {}'.format(best_params_file))
io.log('\nSaving all cross-validation values...')
io.save_object(cv_results, 'intersection_svm_CNNfeatures', ignore=True)
cv_results_file = os.path.abspath(
'./ignore/intersection_svm_CNNfeatures.pickle')
io.log('Saved at {}'.format(cv_results_file))
io.log('\nBEST PARAMS')
io.log('C={} --> accuracy: {:.3f}'.format(best_params['C'], best_accuracy))
def train():
best_accuracy = 0
best_params = {}
cv_results = {}
""" SETTINGS """
settings.n_jobs = 1
# Read the training set
train_images_filenames, train_labels = io.load_training_set()
io.log('Loaded {} train images.'.format(len(train_images_filenames)))
k = 64
io.log('Obtaining dense CNN features...')
start_feature = time.time()
try:
D, L, I = io.load_object('train_CNN_descriptors', ignore=True), \
io.load_object('train_CNN_labels', ignore=True), \
io.load_object('train_CNN_indices', ignore=True)
except IOError:
# load VGG model
base_model = VGG16(weights='imagenet')
# io.save_object(base_model, 'base_model', ignore=True)
# visualize topology in an image
plot(base_model,
to_file='modelVGG16.png',
show_shapes=True,
show_layer_names=True)
# crop the model up to a certain layer
model = Model(input=base_model.input,
output=base_model.get_layer('block5_conv2').output)
D, L, I = feature_extraction.parallel_CNN_features(
train_images_filenames,
train_labels,
model,
num_samples_class=-1,
n_jobs=settings.n_jobs)
io.save_object(D, 'train_CNN_descriptors', ignore=True)
io.save_object(L, 'train_CNN_labels', ignore=True)
io.save_object(I, 'train_CNN_indices', ignore=True)
feature_time = time.time() - start_feature
io.log('Elapsed time: {:.2f} s'.format(feature_time))
for dim_red in pca_reduction:
io.log('Applying PCA ... ')
start_pca = time.time()
settings.pca_reduction = D.shape[1] * dim_red
pca, D_pca = feature_extraction.pca(D)
pca_time = time.time() - start_pca
io.log('Elapsed time: {:.2f} s'.format(pca_time))
for k in codebook_size:
io.log('Creating GMM model (k = {})'.format(k))
start_gmm = time.time()
settings.codebook_size = k
gmm = bovw.create_gmm(
D_pca,
'gmm_{}_pca_{}_CNNfeature'.format(k, settings.pca_reduction))
gmm_time = time.time() - start_gmm
io.log('Elapsed time: {:.2f} s'.format(gmm_time))
io.log('Getting Fisher vectors from training set...')
start_fisher = time.time()
fisher, labels = bovw.fisher_vectors(D_pca,
L,
I,
gmm,
normalization='l2')
fisher_time = time.time() - start_fisher
io.log('Elapsed time: {:.2f} s'.format(fisher_time))
io.log('Scaling features...')
start_scaler = time.time()
std_scaler = StandardScaler().fit(fisher)
vis_words = std_scaler.transform(fisher)
scaler_time = time.time() - start_scaler
io.log('Elapsed time: {:.2f} s'.format(scaler_time))
io.log('Optimizing SVM hyperparameters...')
start_crossvalidation = time.time()
svm = SVC(kernel='precomputed')
random_search = RandomizedSearchCV(svm,
params_distribution,
n_iter=n_iter,
scoring='accuracy',
n_jobs=settings.n_jobs,
refit=False,
cv=3,
verbose=1)
# Precompute Gram matrix
gram = kernels.intersection_kernel(vis_words, vis_words)
random_search.fit(gram, labels)
crossvalidation_time = time.time() - start_crossvalidation
io.log('Elapsed time: {:.2f} s'.format(crossvalidation_time))
# Convert MaskedArrays to ndarrays to avoid unpickling bugs
results = random_search.cv_results_
results['param_C'] = results['param_C'].data
# Appending all parameter-scores combinations
cv_results.update({
(k): {
'cv_results':
results,
'feature_time':
feature_time,
'pca_time':
pca_time,
'gmm_time':
gmm_time,
'fisher_time':
fisher_time,
'scaler_time':
scaler_time,
'crossvalidation_time':
crossvalidation_time,
'total_time':
feature_time + pca_time + gmm_time + fisher_time +
scaler_time + crossvalidation_time
}
})
io.save_object(cv_results,
'intersection_svm_CNNfeatures',
ignore=True)
# Obtaining the parameters which yielded the best accuracy
if random_search.best_score_ > best_accuracy:
best_accuracy = random_search.best_score_
best_params = random_search.best_params_
best_params.update({'k': k, 'pca': dim_red})
io.log('-------------------------------\n')
io.log('\nSaving best parameters...')
io.save_object(best_params,
'best_params_intersection_svm_CNNfeatures',
ignore=True)
best_params_file = os.path.abspath(
'./ignore/best_params_intersection_svm_CNNfeatures.pickle')
io.log('Saved at {}'.format(best_params_file))
io.log('\nSaving all cross-validation values...')
io.save_object(cv_results, 'intersection_svm_CNNfeatures', ignore=True)
cv_results_file = os.path.abspath(
'./ignore/intersection_svm_CNNfeatures.pickle')
io.log('Saved at {}'.format(cv_results_file))
io.log('\nBEST PARAMS')
io.log('k={}, dim_red={}, C={} --> accuracy: {:.3f}'.format(
best_params['k'], best_params['pca'], best_params['C'], best_accuracy))
C=0.0428307453111
pca_reduction = 52
# load VGG model
base_model = VGG16(weights='imagenet')
# crop the model up to a certain layer
model = Model(input=base_model.input, output=base_model.get_layer('block5_conv2').output)
# get train and test images
train_images_filenames = cPickle.load(open('./dataset/train_images_filenames.dat', 'r'))
test_images_filenames = cPickle.load(open('./dataset/test_images_filenames.dat', 'r'))
train_labels = cPickle.load(open('./dataset/train_labels.dat', 'r'))
test_labels = cPickle.load(open('./dataset/test_labels.dat', 'r'))
io.log('\nLoaded {} train images.'.format(len(train_images_filenames)))
io.log('\nLoaded {} test images.'.format(len(test_images_filenames)))
# read and process training images
print 'Getting features from training images'
Train_descriptors = []
Train_label_per_descriptor = []
for i in range(len(train_images_filenames)):
img = image.load_img(train_images_filenames[i], target_size=(224, 224))
x = image.img_to_array(img)
x = np.expand_dims(x, axis=0)
x = preprocess_input(x)
def train():
best_accuracy = 0
best_params = {}
cv_results = {}
""" SETTINGS """
settings.n_jobs = 1
# Read the training set
train_images_filenames, train_labels = io.load_training_set()
io.log('Loaded {} train images.'.format(len(train_images_filenames)))
# Parameter sweep for dense SIFT
for ds in dense_sampling_density:
io.log('Obtaining dense features with sampling parameter {}...'.format(
ds))
start_sift = time.time()
settings.dense_sampling_density = ds
try:
D, L, I = io.load_object('train_dense_descriptors_{}'.format(settings.dense_sampling_density), ignore=True), \
io.load_object('train_dense_labels_{}'.format(settings.dense_sampling_density), ignore=True), \
io.load_object('train_dense_indices_{}'.format(settings.dense_sampling_density), ignore=True)
except IOError:
D, L, I, _ = feature_extraction.parallel_dense(
train_images_filenames,
train_labels,
num_samples_class=-1,
n_jobs=settings.n_jobs)
io.save_object(D,
'train_dense_descriptors_{}'.format(
settings.dense_sampling_density),
ignore=True)
io.save_object(L,
'train_dense_labels_{}'.format(
settings.dense_sampling_density),
ignore=True)
io.save_object(I,
'train_dense_indices_{}'.format(
settings.dense_sampling_density),
ignore=True)
sift_time = time.time() - start_sift
io.log('Elapsed time: {:.2f} s'.format(sift_time))
# Parameter sweep for PCA
for dim_red in pca_reduction:
io.log('Applying PCA (dim = {})...'.format(dim_red))
start_pca = time.time()
settings.pca_reduction = dim_red
pca, D_pca = feature_extraction.pca(D)
pca_time = time.time() - start_pca
io.log('Elapsed time: {:.2f} s'.format(pca_time))
# Parameter sweep for codebook size
for k in codebook_size:
io.log('Creating GMM model (k = {})'.format(k))
start_gmm = time.time()
settings.codebook_size = k
gmm = bovw.create_gmm(
D_pca, 'gmm_{}_dense_{}_pca_{}'.format(k, ds, dim_red))
gmm_time = time.time() - start_gmm
io.log('Elapsed time: {:.2f} s'.format(gmm_time))
io.log('Getting Fisher vectors from training set...')
start_fisher = time.time()
fisher, labels = bovw.fisher_vectors(D_pca,
L,
I,
gmm,
normalization='l2')
fisher_time = time.time() - start_fisher
io.log('Elapsed time: {:.2f} s'.format(fisher_time))
io.log('Scaling features...')
start_scaler = time.time()
std_scaler = StandardScaler().fit(fisher)
vis_words = std_scaler.transform(fisher)
scaler_time = time.time() - start_scaler
io.log('Elapsed time: {:.2f} s'.format(scaler_time))
io.log('Optimizing SVM hyperparameters...')
start_crossvalidation = time.time()
svm = SVC(kernel='precomputed')
random_search = RandomizedSearchCV(svm,
params_distribution,
n_iter=n_iter,
scoring='accuracy',
n_jobs=settings.n_jobs,
refit=False,
cv=3,
verbose=1)
# Precompute Gram matrix
gram = kernels.intersection_kernel(vis_words, vis_words)
random_search.fit(gram, labels)
crossvalidation_time = time.time() - start_crossvalidation
io.log('Elapsed time: {:.2f} s'.format(crossvalidation_time))
# Convert MaskedArrays to ndarrays to avoid unpickling bugs
results = random_search.cv_results_
results['param_C'] = results['param_C'].data
# Appending all parameter-scores combinations
cv_results.update({
(k, dim_red, ds): {
'cv_results':
results,
'sift_time':
sift_time,
'pca_time':
pca_time,
'gmm_time':
gmm_time,
'fisher_time':
fisher_time,
'scaler_time':
scaler_time,
'crossvalidation_time':
crossvalidation_time,
'total_time':
sift_time + pca_time + gmm_time + fisher_time +
scaler_time + crossvalidation_time
}
})
io.save_object(
cv_results,
'intersection_svm_optimization_fisher_vectors_l2',
ignore=True)
# Obtaining the parameters which yielded the best accuracy
if random_search.best_score_ > best_accuracy:
best_accuracy = random_search.best_score_
best_params = random_search.best_params_
best_params.update({'k': k, 'pca': dim_red, 'ds': ds})
io.log('-------------------------------\n')
io.log('\nSaving best parameters...')
io.save_object(
best_params,
'best_params_intersection_svm_optimization_fisher_vectors_l2',
ignore=True)
best_params_file = os.path.abspath(
'./ignore/best_params_intersection_svm_optimization_fisher_vectors_l2.pickle'
)
io.log('Saved at {}'.format(best_params_file))
io.log('\nSaving all cross-validation values...')
io.save_object(cv_results,
'intersection_svm_optimization_fisher_vectors_l2',
ignore=True)
cv_results_file = os.path.abspath(
'./ignore/intersection_svm_optimization_fisher_vectors_l2.pickle')
io.log('Saved at {}'.format(cv_results_file))
io.log('\nBEST PARAMS')
io.log('k={}, C={}, dim_red={}, dense_grid={} --> accuracy: {:.3f}'.format(
best_params['k'], best_params['C'], best_params['pca'],
best_params['ds'], best_accuracy))
def train():
best_accuracy = 0
best_params = {}
cv_results = {}
# load VGG model
base_model = VGG16(weights='imagenet')
# crop the model up to a certain layer
model = Model(input=base_model.input,
output=base_model.get_layer('block5_conv2').output)
# aggregating features with max-pooling
# inputs = Input(shape=[14, 14, 512])
# x = MaxPooling2D((2, 2), strides=(2, 2), name='max_pooling_layer')(inputs)
# model_agg = Model(inputs, x, name='agg_features')
# get train and test images
train_images_filenames = cPickle.load(
open('./dataset/train_images_filenames.dat', 'r'))
train_labels = cPickle.load(open('./dataset/train_labels.dat', 'r'))
io.log('\nLoaded {} train images.'.format(len(train_images_filenames)))
# read and process training images
print 'Getting features from training images'
Train_descriptors = []
Train_label_per_descriptor = []
for i in range(len(train_images_filenames)):
img = image.load_img(train_images_filenames[i], target_size=(224, 224))
x = image.img_to_array(img)
x = np.expand_dims(x, axis=0)
x = preprocess_input(x)
# get the features from images
features_ = model.predict(x)
features = features_[0, :, :, :]
descriptor = features.reshape(features.shape[0] * features.shape[1],
features.shape[2])
# aggregate features
# max value (can be different filters)
#descriptor_agg=descriptor.max(axis=1)
# sum value (of all layers)
#descriptor_agg=np.sum(descriptor,axis=1)
# max value of just one filter
energy = descriptor.max(axis=0)
descriptor_agg = descriptor[:, np.argmax(energy)]
descriptor_agg = np.reshape(descriptor_agg,
[descriptor_agg.shape[0], 1])
Train_descriptors.append(descriptor_agg)
Train_label_per_descriptor.append(train_labels[i])
# Put all descriptors in a numpy array to compute PCA and GMM
size_descriptors = Train_descriptors[0].shape[1]
#size_descriptors=1
Desc = np.zeros(
(np.sum([len(p) for p in Train_descriptors]), size_descriptors),
dtype=np.uint8)
startingpoint = 0
for i in range(len(Train_descriptors)):
Desc[startingpoint:startingpoint +
len(Train_descriptors[i])] = Train_descriptors[i]
startingpoint += len(Train_descriptors[i])
for k in codebook_size:
print('Computing gmm with ' + str(k) + ' centroids')
gmm = ynumpy.gmm_learn(np.float32(Desc), k)
# io.save_object(gmm, 'gmm_NN_agg_features_max')
# Compute the fisher vectors of the training images
print('Computing fisher vectors')
fisher = np.zeros((len(Train_descriptors), k * 1 * 2),
dtype=np.float32)
for i in xrange(len(Train_descriptors)):
descriptor = Train_descriptors[i]
# descriptor = np.float32(pca.transform(descriptor))
aux = ynumpy.fisher(gmm, descriptor, include=['mu', 'sigma'])
fisher[i, :] = np.reshape(aux, [1, aux.shape[0]])
#fisher[i,:]=aux
# L2 normalization - reshape to avoid deprecation warning, checked that the result is the same
fisher[i, :] = preprocessing.normalize(fisher[i, :].reshape(1, -1),
norm='l2')
# CV in SVM training
io.log('Scaling features...')
std_scaler = StandardScaler().fit(fisher)
vis_words = std_scaler.transform(fisher)
io.log('Optimizing SVM hyperparameters...')
svm = SVC(kernel='precomputed')
random_search = RandomizedSearchCV(svm,
params_distribution,
n_iter=n_iter,
scoring='accuracy',
n_jobs=1,
refit=False,
cv=3,
verbose=1)
# Precompute Gram matrix
gram = kernels.intersection_kernel(vis_words, vis_words)
random_search.fit(gram, train_labels)
# Convert MaskedArrays to ndarrays to avoid unpickling bugs
results = random_search.cv_results_
results['param_C'] = results['param_C'].data
# Appending all parameter-scores combinations
cv_results.update({(k): {
'cv_results': results,
}})
io.save_object(cv_results,
'intersection_svm_CNNfeatures_aggregate_energy',
ignore=True)
# Obtaining the parameters which yielded the best accuracy
if random_search.best_score_ > best_accuracy:
best_accuracy = random_search.best_score_
best_params = random_search.best_params_
best_params.update({'k': k})
io.log('-------------------------------\n')
io.log('\nSaving best parameters...')
io.save_object(best_params,
'best_params_intersection_svm_CNNfeatures_aggregate_energy',
ignore=True)
best_params_file = os.path.abspath(
'./ignore/best_params_intersection_svm_CNNfeatures_aggregate_energy.pickle'
)
io.log('Saved at {}'.format(best_params_file))
io.log('\nSaving all cross-validation values...')
io.save_object(cv_results,
'intersection_svm_CNNfeatures_aggregate_energy',
ignore=True)
cv_results_file = os.path.abspath(
'./ignore/intersection_svm_CNNfeatures_aggregate_energy.pickle')
io.log('Saved at {}'.format(cv_results_file))
io.log('\nBEST PARAMS')
io.log('k={}, C={} --> accuracy: {:.3f}'.format(best_params['k'],
best_params['C'],
best_accuracy))
