def reranking(Q, data, inds, names, top_k = 50):
vecs_sum = data[0, :]
for i in range(1, top_k):
vecs_sum += data[inds[i], :]
vec_mean = vecs_sum/float(top_k)
Q = normalize(Q - vec_mean)
for i in range(top_k):
data[i, :] = normalize(data[i, :] - vec_mean)
sub_data = data[:top_k]
sub_idxs, sub_rerank_dists, sub_rerank_names = compute_cosin_distance(Q, sub_data, names[:top_k])
names[:top_k] = sub_rerank_names
return names
def simple_query_expansion(Q, data, inds, top_k=5):
for i in range(top_k):
temp = np.zeros(Q.shape)
# print(data.shape)
for k, j in enumerate(inds.T[i]):
temp[k] = data[j]
Q += (1.0 * (top_k - i) / float(top_k)) * temp
return normalize(Q)
def extract_crow_feature(img, cls_model):
output = cls_model(img)
output_cpu = output.cpu().data.numpy()
X = output_cpu.squeeze()
f_1 = crow.normalize(crow.apply_crow_aggregation(X)).flatten()
return f_1
def simple_query_expansion(Q, data, inds, top_k=10):
"""
Get the top-k closest vectors, average and re-query
:param ndarray Q:
query vector
:param ndarray data:
index data vectors
:param ndarray inds:
the indices of index vectors in ascending order of distance
:param int top_k:
the number of closest vectors to consider
:returns ndarray idx:
the indices of index vectors in ascending order of distance
:returns ndarray dists:
the squared distances
"""
Q += data[inds[:top_k],:].sum(axis=0)
return normalize(Q)
def simple_query_expansion(Q, data, inds, top_k=10):
"""
Get the top-k closest vectors, average and re-query
:param ndarray Q:
query vector
:param ndarray data:
index data vectors
:param ndarray inds:
the indices of index vectors in ascending order of distance
:param int top_k:
the number of closest vectors to consider
:returns ndarray idx:
the indices of index vectors in ascending order of distance
:returns ndarray dists:
the squared distances
"""
Q += data[inds[:top_k], :].sum(axis=0)
return normalize(Q)
def extract_feat_3_yalers(img_path):
model =load_model(dataPath+'/f1cn_model.49-1.613893.hdf5')
layer_1 = K.function([model.layers[0].input], [model.layers[7].output])
img = image.load_img(img_path, target_size=(224,224))
img = image.img_to_array(img)
img = np.expand_dims(img, axis=0)
img = preprocess_input(img)
f1 = layer_1([img])[0]
feat_5=convert_kernel(f1[0].T)
feature_crow_5=apply_crow_aggregation(feat_5)
feature_norm_5=normalize(feature_crow_5)
return np.sqrt(feature_norm_5)
def load_face_2(face_data1, face_data2):
face_dict = {}
movie_list = []
for movie, info in face_data1.items():
movie_list.append(movie)
casts = info['cast']
candidates = info['candidates']
cast_ids, cast_ffeats = [], []
for index, cast in enumerate(casts):
feat1 = cast['ffeat']
feat2 = face_data2[movie]['cast'][index]['ffeat']
assert cast['id'] == face_data2[movie]['cast'][index]['id']
feat = np.hstack((feat1 * 0.5, feat2 * 0.5))
# feat = normalize(feat)
cast_ffeats.append(feat)
cast_ids.append(cast['id'])
cast_ffeats = np.array(cast_ffeats)
candi_f_ids, candi_f_ffeats = [], []
for index, candidate in enumerate(candidates):
if candidate['ffeat'] is not None:
feat1 = candidate['ffeat']
feat2 = face_data2[movie]['candidates'][index]['ffeat']
assert candidate['id'] == face_data2[movie]['candidates'][
index]['id']
feat = np.hstack((feat1, feat2))
feat = normalize(feat)
candi_f_ids.append(candidate['id'])
candi_f_ffeats.append(feat)
candi_f_ffeats = np.array(candi_f_ffeats)
face_dict.update({
movie: {
'cast_ids': cast_ids,
'cast_ffeats': cast_ffeats,
'candi_f_ids': candi_f_ids,
'candi_f_ffeats': candi_f_ffeats,
}
})
return face_dict, movie_list
def simple_query_expansion(Q, data, inds, top_k=10):
Q += data[inds[:top_k], :].sum(axis=0)
return normalize(Q)
def query_images(groundtruth_dir, image_dir, dataset, cropped=True):
"""
Extract features from the Oxford or Paris dataset.
:param str groundtruth_dir:
the directory of the groundtruth files (which includes the query files)
:param str image_dir:
the directory of dataset images
:param str dataset:
the name of the dataset, either 'oxford' or 'paris'
:param bool cropped:
flag to optionally disable cropping
:yields Image img:
the Image object
:yields str query_name:
the name of the query
"""
imgs = []
query_names = []
fake_query_names = []
feats_crop = []
modelDir = "/home/yuanyong/py/crow_retrieval/model"
MODEL = "vgg.model"
PROTO = "pool5.prototxt"
caffemodel = os.path.join(modelDir, MODEL)
prototxt = os.path.join(modelDir, PROTO)
# set gpu card
layer = 'pool5'
caffe.set_device(6)
caffe.set_mode_gpu()
# init NN
net = caffe.Net(prototxt, caffemodel, caffe.TEST)
net.forward()
for f in glob.iglob(os.path.join(groundtruth_dir, '*_query.txt')):
fake_query_name = os.path.splitext(os.path.basename(f))[0].replace('_query', '')
fake_query_names.append(fake_query_name)
query_name, x, y, w, h = open(f).read().strip().split(' ')
if dataset == 'oxford':
query_name = query_name.replace('oxc1_', '')
query_names.append('%s.jpg' % query_name)
img = cv2.imread(os.path.join(image_dir, '%s.jpg' % query_name), 1) # BGR
if cropped:
x, y, w, h = map(float, (x, y, w, h))
x, y, w, h = map(lambda d: int(round(d)), (x, y, w, h))
else:
x, y, w, h = (0, 0, img.shape[1], img.shape[0])
img = img[y:y+h, x:x+w]
d = np.float32(img)
# VggNet
d -= np.array((104.00698793, 116.66876762, 122.67891434))
d = d.transpose((2, 0, 1))
feat = extract_raw_features(net, layer, d)
#feat = extract_multi_raw_features(net, layer, d)
feat = apply_crow_aggregation(feat)
# L2-normalize feature
feat = normalize(feat, copy=False)
feats_crop.append(feat)
imgs.append(img)
return imgs, feats_crop, query_names, fake_query_names
do_QE = True
topK = 10
do_crop = True
do_pca = True
do_rerank = False
redud_d = 256
# load all features
start = timeit.default_timer()
files = glob.glob(feats_files)
feats, names = load_files(files)
stop = timeit.default_timer()
print "load time: %f seconds\n" % (stop - start)
# L2-normalize features
feats = normalize(feats, copy=False)
# PCA reduce dimension
if do_pca:
#import pickle
#whitening_params = {}
#if os.path.isfile('../model/pca_model.pkl'):
# with open( '../model/pca_model.pkl' , 'rb') as f:
# whitening_params['pca'] = pickle.load(f)
_, whitening_params = run_feature_processing_pipeline(feats, d=redud_d, copy=True)
feats, _ = run_feature_processing_pipeline(feats, params=whitening_params)
imgs, query_feats, query_names, fake_query_names = query_images(gt_files, dir_images, 'oxford', do_crop)
#print query_names
aps = []
def extract_feat(self, img_path):
"""
img = image.load_img(img_path, target_size=(self.input_shape[0], self.input_shape[1]))
img = image.img_to_array(img)
"""
img = image.load_img(img_path)
img = image.img_to_array(img)
h, w, c = img.shape
resize_h=h
resize_w=w
minlength=min(h,w)
if minlength>224:
beta=minlength/224
resize_h=int(h/beta)
resize_w=int(w/beta)
img=cv2.resize(img,(resize_h,resize_w))
img = np.expand_dims(img, axis=0)
img = preprocess_input(img)
feat_0 = self.model_vgg_block1_conv2.predict(img)
feat_0=convert_kernel(feat_0[0].T)
feature_crow_0=apply_crow_aggregation(feat_0)
feature_norm_0=normalize(feature_crow_0)
feature_mean_norm_0=normalize(preprocessing.scale(feature_crow_0,axis=0, with_mean=True, with_std=False, copy=True))
feat_1 = self.model_vgg_block2_conv2.predict(img)
feat_1=convert_kernel(feat_1[0].T)
feature_crow_1=apply_crow_aggregation(feat_1)
feature_norm_1=normalize(feature_crow_1)
feature_mean_norm_1=normalize(preprocessing.scale(feature_crow_1,axis=0, with_mean=True, with_std=False, copy=True))
feat_2= self.model_vgg_block3_conv1.predict(img)
feat_2=convert_kernel(feat_2[0].T)
feature_crow_2=apply_crow_aggregation(feat_2)
feature_norm_2=normalize(feature_crow_2)
feature_mean_norm_2=normalize(preprocessing.scale(feature_crow_2,axis=0, with_mean=True, with_std=False, copy=True))
feature_448=np.hstack((np.hstack((feature_crow_0.T,feature_crow_1.T)),feature_crow_2.T))
feature_448_norm=np.hstack((np.hstack((feature_norm_0.T,feature_norm_1.T)),feature_norm_2.T))
feature_448_mean_norm=np.hstack((np.hstack((feature_mean_norm_0.T,feature_mean_norm_1.T)),feature_mean_norm_2.T))
feat_3 = self.model_vgg_block3_conv3.predict(img)
feat_3=convert_kernel(feat_3[0].T)
feature_crow_3=apply_crow_aggregation(feat_3)
feature_norm_3=normalize(feature_crow_3)
feature_mean_norm_3=normalize(preprocessing.scale(feature_crow_3,axis=0, with_mean=True, with_std=False, copy=True))
feat_4 = self.model_vgg_block4_conv3.predict(img)
feat_4=convert_kernel(feat_4[0].T)
feature_crow_4=apply_crow_aggregation(feat_4)
feature_norm_4=normalize(feature_crow_4)
feature_mean_norm_4=normalize(preprocessing.scale(feature_crow_4,axis=0, with_mean=True, with_std=False, copy=True))
feat_5= self.model_vgg_block5_conv3.predict(img)
feat_5=convert_kernel(feat_5[0].T)
feature_crow_5=apply_crow_aggregation(feat_5)
feature_norm_5=normalize(feature_crow_5)
feature_mean_norm_5=normalize(preprocessing.scale(feature_crow_5,axis=0, with_mean=True, with_std=False, copy=True))
feature_1280=np.hstack((np.hstack((feature_crow_3.T,feature_crow_4.T)),feature_crow_5.T))
feature_1280_norm=np.hstack((np.hstack((feature_norm_3.T,feature_norm_4.T)),feature_norm_5.T))
feature_1280_mean_norm=np.hstack((np.hstack((feature_mean_norm_3.T,feature_mean_norm_4.T)),feature_mean_norm_5.T))
#print(feature_norm.shape)
#feature,pca_prams=run_feature_processing_pipeline(feature_norm)
return np.hstack((feature_448.T,feature_1280.T)),np.hstack((feature_448_norm.T,feature_1280_norm.T)),np.hstack((feature_448_mean_norm.T,feature_1280_mean_norm.T))
