def main():
sys.path.append(os.path.join(os.path.dirname(__file__), '..'))
from keras_video_classifier.library.recurrent_networks import VGG16BidirectionalLSTMVideoClassifier
#from keras_video_classifier.library.utility.ucf.UCF101_loader import load_ucf, scan_ucf_with_labels
from keras_video_classifier.library.utility.ucf.UCF101_loader import scan_ucf_with_labels, scan_ucf_predict
vgg16_include_top = True
data_dir_path = os.path.join(os.path.dirname(__file__), 'very_large_data')
model_dir_path = os.path.join(os.path.dirname(__file__), 'models',
'videos')
config_file_path = VGG16BidirectionalLSTMVideoClassifier.get_config_file_path(
model_dir_path, vgg16_include_top=vgg16_include_top)
weight_file_path = VGG16BidirectionalLSTMVideoClassifier.get_weight_file_path(
model_dir_path, vgg16_include_top=vgg16_include_top)
np.random.seed(42)
#load_ucf(data_dir_path)
predictor = VGG16BidirectionalLSTMVideoClassifier()
predictor.load_model(config_file_path, weight_file_path)
#videos = scan_ucf_with_labels(data_dir_path, [label for (label, label_index) in predictor.labels.items()])
videos = scan_ucf_predict(data_dir_path)
video_file_path_list = np.array(
[video_file_path for video_file_path in videos.keys()])
np.random.shuffle(video_file_path_list)
correct_count = 0
count = 0
arr = []
for video_file_path in video_file_path_list:
#label = videos[video_file_path]
video_number = video_file_path[video_file_path.rfind('/') +
1:video_file_path.rfind('.')]
video_number = int(video_number)
predicted_label = predictor.predict(video_file_path)
#print('predicted: ' + predicted_label + ' actual: ' + label)
#correct_count = correct_count + 1 if label == predicted_label else correct_count
count += 1
#print('predicted: ' + predicted_label)
if predicted_label == 'smoking':
arr.append(video_number)
#accuracy = correct_count*100 / count
#print('accuracy: ', accuracy)
arr.sort(key=lambda x: x)
print('The timestamps are:\n')
for num in arr:
print(str(5 * num) + ' -> ' + str(5 * num + 5) + '\n')
def main():
K.set_image_dim_ordering('tf')
sys.path.append(os.path.join(os.path.dirname(__file__), '..'))
from keras_video_classifier.library.recurrent_networks import VGG16BidirectionalLSTMVideoClassifier
from keras_video_classifier.library.utility.plot_utils import plot_and_save_history
from keras_video_classifier.library.utility.ucf.UCF101_loader import load_ucf
data_set_name = 'UCF-101'
input_dir_path = os.path.join(os.path.dirname(__file__), 'very_large_data')
output_dir_path = os.path.join(os.path.dirname(__file__), 'models',
data_set_name)
report_dir_path = os.path.join(os.path.dirname(__file__), 'reports',
data_set_name)
np.random.seed(42)
# this line downloads the video files of UCF-101 dataset if they are not available in the very_large_data folder
load_ucf(input_dir_path)
classifier = VGG16BidirectionalLSTMVideoClassifier()
history = classifier.fit(data_dir_path=input_dir_path,
model_dir_path=output_dir_path,
vgg16_include_top=False,
data_set_name=data_set_name)
plot_and_save_history(
history, VGG16BidirectionalLSTMVideoClassifier.model_name,
report_dir_path + '/' +
VGG16BidirectionalLSTMVideoClassifier.model_name +
'-hi-dim-history.png')
def classify(file_path):
# sys.path.append(os.path.join(os.path.dirname(__file__), '..'))
from keras_video_classifier.library.recurrent_networks import VGG16BidirectionalLSTMVideoClassifier
vgg16_include_top = True
model_dir_path = os.path.join(os.path.dirname(__file__), 'models/dataset')
config_file_path = VGG16BidirectionalLSTMVideoClassifier.get_config_file_path(model_dir_path,
vgg16_include_top=vgg16_include_top)
weight_file_path = VGG16BidirectionalLSTMVideoClassifier.get_weight_file_path(model_dir_path,
vgg16_include_top=vgg16_include_top)
np.random.seed(42)
predictor = VGG16BidirectionalLSTMVideoClassifier()
predictor.load_model(config_file_path, weight_file_path)
predicted_label, face_images = predictor.predict(file_path)
return predicted_label, face_images
def main():
sys.path.append(os.path.join(os.path.dirname(__file__), '..'))
from keras_video_classifier.library.recurrent_networks import VGG16BidirectionalLSTMVideoClassifier
from keras_video_classifier.library.utility.ucf.UCF101_loader import load_ucf, scan_ucf_with_labels
vgg16_include_top = True
data_dir_path = os.path.join(os.path.dirname(__file__), 'very_large_data')
model_dir_path = os.path.join(os.path.dirname(__file__), 'models',
'UCF-101')
config_file_path = VGG16BidirectionalLSTMVideoClassifier.get_config_file_path(
model_dir_path, vgg16_include_top=vgg16_include_top)
weight_file_path = VGG16BidirectionalLSTMVideoClassifier.get_weight_file_path(
model_dir_path, vgg16_include_top=vgg16_include_top)
np.random.seed(42)
load_ucf(data_dir_path)
predictor = VGG16BidirectionalLSTMVideoClassifier()
predictor.load_model(config_file_path, weight_file_path)
print('reaching here three')
videos = scan_ucf_with_labels(
data_dir_path,
[label for (label, label_index) in predictor.labels.items()])
video_file_path_list = np.array([file_path for file_path in videos.keys()])
np.random.shuffle(video_file_path_list)
correct_count = 0
count = 0
for video_file_path in video_file_path_list:
label = videos[video_file_path]
predicted_label = predictor.predict(video_file_path)
print('predicted: ' + predicted_label + ' actual: ' + label)
correct_count = correct_count + 1 if label == predicted_label else correct_count
count += 1
accuracy = correct_count / count
print('correct_count: ' + str(correct_count) + ' count: ' + str(count))
print('accuracy: ', accuracy)
def main():
sys.path.append(os.path.join(os.path.dirname(__file__), '..'))
from keras_video_classifier.library.recurrent_networks import VGG16BidirectionalLSTMVideoClassifier
from keras_video_classifier.library.utility.ucf.UCF101_loader import load_ucf, scan_ucf_with_labels
vgg16_include_top = False
data_dir_path = os.path.join(os.path.dirname(__file__), 'very_large_data')
model_dir_path = os.path.join(os.path.dirname(__file__), 'models/giri')
config_file_path = VGG16BidirectionalLSTMVideoClassifier.get_config_file_path(
model_dir_path, vgg16_include_top=vgg16_include_top)
weight_file_path = VGG16BidirectionalLSTMVideoClassifier.get_weight_file_path(
model_dir_path, vgg16_include_top=vgg16_include_top)
np.random.seed(42)
load_ucf(data_dir_path)
predictor = VGG16BidirectionalLSTMVideoClassifier()
predictor.load_model(config_file_path, weight_file_path)
videos = scan_ucf_with_labels(
data_dir_path,
[label for (label, label_index) in predictor.labels.items()])
video_file_path_list = np.array([file_path for file_path in videos.keys()])
np.random.shuffle(video_file_path_list)
correct_count = 0
count = 0
for video_file_path in video_file_path_list:
label = videos[video_file_path]
correct_count, count = predictor.predict(video_file_path, label,
correct_count, count)
def main():
sys.path.append(os.path.join(os.path.dirname(__file__), '..'))
from keras_video_classifier.library.recurrent_networks import VGG16BidirectionalLSTMVideoClassifier
from keras_video_classifier.library.utility.crime.UCF_Crime_loader import load_ucf, scan_ucf_with_labels
vgg16_include_top = False
data_dir_path = os.path.join(os.path.dirname(__file__), 'very_large_data')
model_dir_path = os.path.join(os.path.dirname(__file__), 'models',
'UCF-Anomaly-Detection-Dataset')
config_file_path = VGG16BidirectionalLSTMVideoClassifier.get_config_file_path(
model_dir_path, vgg16_include_top=vgg16_include_top)
weight_file_path = VGG16BidirectionalLSTMVideoClassifier.get_weight_file_path(
model_dir_path, vgg16_include_top=vgg16_include_top)
np.random.seed(42)
load_ucf(data_dir_path)
predictor = VGG16BidirectionalLSTMVideoClassifier()
predictor.load_model(config_file_path, weight_file_path)
videos = scan_ucf_with_labels(
data_dir_path,
[label for (label, label_index) in predictor.labels.items()])
print("PREDICT_VIDEOS", videos)
video_file_path_list = np.array([file_path for file_path in videos.keys()])
print("video_file_path_list:", video_file_path_list)
np.random.shuffle(video_file_path_list)
correct_count = 0
count = 0
top5_correct_count = 0
print("Begin Predict Using UCF-CRIME-Model: ")
print("Video_file_path_list: ", video_file_path_list)
print(len(video_file_path_list))
for video_file_path in video_file_path_list:
label = videos[video_file_path]
predicted_label = predictor.predict_top5(video_file_path)
print('Top 1 predicted: ' + str(predicted_label[-1]) + ' actual: ' +
label)
correct_count = correct_count + 1 if label == predicted_label[
-1] else correct_count
count += 1
accuracy = correct_count / count
print('Top 5 predicted: ' + str(predicted_label) + ' actual: ' + label)
top5_correct_count = top5_correct_count + 1 if [
s for s in predicted_label if label in s
] else top5_correct_count
accuracy_5 = top5_correct_count / count
print('correct_count: ' + str(correct_count) + ' count: ' + str(count))
print('top5_correct_count: ' + str(top5_correct_count) + ' count: ' +
str(count))
print('accuracy: ', accuracy)
print('accuracy_top_5: ', accuracy_5)
def Detect(self, fullname):
sys.path.append(os.path.join(os.path.dirname(__file__), '..'))
from keras_video_classifier.library.recurrent_networks import VGG16BidirectionalLSTMVideoClassifier
from keras_video_classifier.library.utility.ucf.UCF101_loader import load_ucf, scan_ucf_with_labels
vgg16_include_top = True
data_dir_path = os.path.join(os.path.dirname(__file__),
'very_large_data')
model_dir_path = os.path.join(os.path.dirname(__file__), 'models',
'UCF-101')
demo_dir_path = os.path.join(os.path.dirname(__file__), 'real-data')
config_file_path = VGG16BidirectionalLSTMVideoClassifier.get_config_file_path(
model_dir_path, vgg16_include_top=vgg16_include_top)
weight_file_path = VGG16BidirectionalLSTMVideoClassifier.get_weight_file_path(
model_dir_path, vgg16_include_top=vgg16_include_top)
np.random.seed(42)
load_ucf(data_dir_path)
predictor = VGG16BidirectionalLSTMVideoClassifier()
predictor.load_model(config_file_path, weight_file_path)
print('reaching here three')
unusual_actions = []
usual_actions = []
interval = 3
txtFile = open('./reports/prediction_Each_Interval_Report' + '.txt',
'w')
if os.path.isfile(fullname):
print("Predicting video: {}".format(os.path.basename(fullname)))
predicted_labels = predictor.predict(fullname, interval=interval)
print("TOTAL {} ACTIONS".format(len(predicted_labels)))
for index in range(len(predicted_labels)):
if (predicted_labels[index] == "Unusual action"):
time = str(datetime.timedelta(seconds=index))
print('predicted: ' + predicted_labels[index] +
' at: {0}'.format(time))
unusual_actions.append(list((fullname, time)))
elif (predicted_labels[index] == "Usual action"):
time = str(datetime.timedelta(seconds=index))
print('predicted: ' + predicted_labels[index] +
' at: {0}'.format(time))
usual_actions.append(list((fullname, time)))
print("TOTAL: {} unusual actions and {} usual actions".format(
len(unusual_actions), len(usual_actions)))
txtFile.write("Unsual actions")
txtFile.write('\n')
for item in unusual_actions:
txtFile.write(item[0] + ' at ' + item[1])
txtFile.write('\n')
txtFile.write("Usual actions")
txtFile.write('\n')
for item in usual_actions:
txtFile.write(item[0] + ' at ' + item[1])
txtFile.write('\n')
txtFile.close()
self.SplitVideo(fullname, unusual_actions)
return self.DrawToVideo(fullname, unusual_actions)