def print_matrix(mat):
for i in range(len(mat)):
for j in range(len(mat[i])):
if (j + 1 != len(mat[i])):
stdout.write(str(precision(mat[i][j])) + "\t")
else:
stdout.write(str(precision(mat[i][j])))
stdout.write("\n")
def f1_score(y_true, y_pred):
pr = precision(y_true, y_pred)
re = recall(y_true, y_pred)
return 2 * pr * re / (pr + re)
#precision&map&p@k
sum_ap=0
sum_p5=0
sum_p20=0
for q in results:
rele_set=set([])
if q in relevant:
rele_set=relevant[q]
#if there is no relevant documents for current query, exclude it
else:
validQ_Count-=1
continue
table_path=p_dir+'/Q'+q+".txt"
checkDup(table_path)
[precision_table,ap]=precision(rele_set,results[q],table_path)
sum_p5+=precision_table[4]
sum_p20+=precision_table[19]
sum_ap+=ap
fd_pre.write(q)
for x in precision_table:
fd_pre.write("\t"+str(x))
fd_pre.write('\n')
fd_pre.close()
AP5=sum_p5/validQ_Count
AP20=sum_p20/validQ_Count
MAP=sum_ap/validQ_Count
#recall&mrr
sum_rr=0
r_dir=recall_table_dir+'/'+system
def run():
use_gpu = 1
# top K returned images
top_k = 1000
feat_len = 48 # number of activation output on hidden layer
# Absolute path of this project
root_dir = '/opt/work/caffe-cvprw15'
# set result folder
result_folder = os.path.join(root_dir, 'analysis')
# models
model_file = os.path.join(
root_dir, 'examples/cvprw15-cifar10/KevinNet_CIFAR10_48.caffemodel')
model_def_file = os.path.join(
root_dir,
'examples/cvprw15-cifar10/KevinNet_CIFAR10_48_deploy.prototxt')
# train-test
test_file_list = os.path.join(
root_dir, 'examples/cvprw15-cifar10/dataset/test-file-list.txt')
test_label_file = os.path.join(
root_dir, 'examples/cvprw15-cifar10/dataset/test-label.txt')
train_file_list = os.path.join(
root_dir, 'examples/cvprw15-cifar10/dataset/train-file-list.txt')
train_label_file = os.path.join(
root_dir, 'examples/cvprw15-cifar10/dataset/train-label.txt')
# outputs
#feat_test_file = '{:s}/feat-test.mat'.format(result_folder)
#feat_train_file = '{:s}/feat-train.mat'.format(result_folder)
binary_test_file = '{:s}/binary-test.mat'.format(result_folder)
binary_train_file = '{:s}/binary-train.mat'.format(result_folder)
py_feat_test_file = '{:s}/feat-test.pkl'.format(result_folder)
py_feat_train_file = '{:s}/feat-train.pkl'.format(result_folder)
py_binary_test_file = '{:s}/binary-test.pkl'.format(result_folder)
py_binary_train_file = '{:s}/binary-train.pkl'.format(result_folder)
USE_MAT_FEAT = False # whether load MATLAB code generated and saved feature files, or not
# feature extraction- test set
if USE_MAT_FEAT and os.path.exists(binary_test_file): # load .mat file
data = h5py.File(binary_test_file)
binary_test = data['binary_test'][:] # 10000 x 48
#binary_test = binary_test.T # transpose 转置 => 48 x 1000
elif os.path.exists(py_binary_test_file):
fid = open(py_binary_test_file, 'rb')
binary_test = cPickle.load(fid)
fid.close()
else:
feat_test = pycaffe_batch_feat(test_file_list, use_gpu, feat_len,
model_def_file, model_file, root_dir)
binary_test = (feat_test > 0.5).astype(int)
# save extracted features to disk
with open(py_feat_test_file, 'wb') as fid:
cPickle.dump(feat_test, fid)
with open(py_binary_test_file, 'wb') as fid:
cPickle.dump(binary_test, fid)
# feature extraction- train set
if USE_MAT_FEAT and os.path.exists(binary_train_file): # load .mat file
data = h5py.File(binary_train_file)
binary_train = data['binary_train'][:] # 10000 x 48
#binary_train = binary_train.T # transpose 转置 => 48 x 1000
elif os.path.exists(py_binary_train_file):
fid = open(py_binary_train_file, 'rb')
binary_train = cPickle.load(fid)
fid.close()
else:
feat_train = pycaffe_batch_feat(train_file_list, use_gpu, feat_len,
model_def_file, model_file, root_dir)
binary_train = (feat_train > 0.5).astype(int)
# save extracted features to disk
with open(py_feat_train_file, 'wb') as fid:
cPickle.dump(feat_train, fid)
with open(py_binary_train_file, 'wb') as fid:
cPickle.dump(binary_train, fid)
# load training images' labels. double type
trn_label = load_label(train_label_file)
tst_label = load_label(test_label_file)
mAP, precision_at_k = precision(trn_label, binary_train, tst_label,
binary_test, top_k, 1)
print('mAP = {:f}'.format(mAP))
saved_result = {'mAP': mAP, 'precision_at_k': precision_at_k}
save_file = os.path.join(root_dir, 'saved_result.pkl')
fid = open(save_file, 'wb')
cPickle.dump(saved_result, fid)
fid.close()
def run():
use_gpu = 1
# top K returned images
top_k = 1000
feat_len = 48 # number of activation output on hidden layer
# Absolute path of this project
root_dir = 'E:/GitHub/cvprw15-win/'
root_dataset = 'E:/GitHub/dataset/cifar10/'
# set result folder
result_folder = os.path.join(root_dir, 'analysis/cvprw15-cifar10')
# models
model_file = os.path.join(
root_dir, 'examples/cvprw15-cifar10/KevinNet_CIFAR10_48.caffemodel')
model_def_file = os.path.join(
root_dir,
'examples/cvprw15-cifar10/KevinNet_CIFAR10_48_deploy.prototxt')
# train-test
test_file_list = os.path.join(
root_dataset, 'examples/cvprw15-cifar10/dataset/test-file-list.txt')
test_label_file = os.path.join(
root_dataset, 'examples/cvprw15-cifar10/dataset/test-label.txt')
train_file_list = os.path.join(
root_dataset, 'examples/cvprw15-cifar10/dataset/train-file-list.txt')
train_label_file = os.path.join(
root_dataset, 'examples/cvprw15-cifar10/dataset/train-label.txt')
# outputs
feat_test_file = '{:s}/feat-test.mat'.format(result_folder)
feat_train_file = '{:s}/feat-train.mat'.format(result_folder)
binary_test_file = '{:s}/binary-test.mat'.format(result_folder)
binary_train_file = '{:s}/binary-train.mat'.format(result_folder)
# feature extraction- test set
if os.path.exists(binary_test_file): # load .mat file
binary_test = scio.loadmat(binary_test_file)['binary_test'].astype(
float)
binary_test = binary_test.T
else:
feat_test, _ = pycaffe_batch_feat(test_file_list, use_gpu, feat_len,
model_def_file, model_file, root_dir)
scio.savemat(feat_test_file, {'feat_test': feat_test.T})
binary_test = (feat_test > 0.5).astype(int)
scio.savemat(binary_test_file, {'binary_test': binary_test.T})
# feature extraction- train set
if os.path.exists(binary_train_file): # load .mat file
binary_train = scio.loadmat(binary_train_file)['binary_train'].astype(
float)
binary_train = binary_train.T
else:
feat_train, _ = pycaffe_batch_feat(train_file_list, use_gpu, feat_len,
model_def_file, model_file,
root_dir)
scio.savemat(feat_train_file, {'feat_train': feat_train.T})
binary_train = (feat_train > 0.5).astype(int)
scio.savemat(binary_train_file, {'binary_train': binary_train.T})
# load training images' labels. double type
trn_label = load_label(train_label_file)
tst_label = load_label(test_label_file)
mAP, precision_at_k = precision(trn_label, binary_train, tst_label,
binary_test, top_k, 1)
print('mAP = {:f}'.format(mAP))
saved_result = {'mAP': mAP, 'precision_at_k': precision_at_k}
save_file = os.path.join(result_folder, 'saved_result.pkl')
fid = open(save_file, 'wb')
pickle.dump(saved_result, fid)
fid.close()