def predictVideoGeneral(video_file, video_path, name_model, save_path, model_path, model_folder, img_height, img_width):
video_title = os.path.splitext(video_file)[0]
#img_size = 90
vp.extractFaceImages(video_file, video_path)
face_images = vp.loadFaceImages(video_file,video_path)
labels, confidence = vp.predictVideo(face_images,name_model,img_height,img_width, DEFAULT_EXPR, model_path, model_folder) # Predict the facial expression label for each of the face images (that is, for each video frame)
#print 'Predicted labels: '+str(labels)
clean_labels = vp.cleanLabels(labels)
#print 'Cleaned labels: '+str(clean_labels)
perframe_labels_areStored = vp.savePredictionsPerFrame(clean_labels, save_path + video_title + '_predictions_perframe.csv')
print perframe_labels_areStored
perexp_labels_areStored = vp.savePredictionsPerExpression(clean_labels, save_path + video_title + 'predictions_perexp.csv')
print perexp_labels_areStored
# Play the tested video with the labels written in text format over the image
vp.playLabeledVideo(video_path,video_file, clean_labels, True)
# Clean all the files created by OpenFace
vproc.deleteProcessData(video_path)
return clean_labels
def confidenceLevelsVideoSetDefault(videos_path,nmodel):
# Get the list of names of the video files
classes = ['concerned','enthusiastic','neutral']
model_path = 'defaultmodels/'
img_height = 90
img_width = 90
videosList = os.listdir(videos_path) # Lists all files (and directories) in the folder
confidence_videoset = []
for video in videosList:
if os.path.isfile(os.path.join(videos_path, video)): # Considers files only
vp.extractFaceImages(video, videos_path)
for video in videosList:
if os.path.isfile(os.path.join(videos_path, video)):
#extractFaceImages(video, videos_path)
face_images = vp.loadFaceImages(video,videos_path)
print 'Current video: '+str(video)
labels, confidence = vp.predictVideo(face_images,'face-exp-model'+str(nmodel),img_height,img_width, DEFAULT_EXPR, model_path, models[nmodel]) # Run the model
confidence_videoset.append([video]+confidence) # Store the mean and standard deviation of the per-frame confidence levels, representative of confidence of full video
sorted(confidence_videoset, key=lambda x: x[1]) # Sort them per ascending mean value
# The video(s) with the lowest confidence level are considered relevant for the model
# (and hence will later be annotated and the model re-trained)
for video in confidence_videoset: # Print the confidence levels of all videos, in ascending order
print str(video[0])+" -- "+"mean cnf: "+str(video[1])+", stddev conf: "+str(video[2])
# Clean all the files created by OpenFace
vproc.deleteProcessData(videos_path)
def videoSetProcessing(self, annot_type):
if annot_type==0: # MMI-based annotations format
vproc.videoProcessorMMI(self._data_path, self._list_classes)
elif annot_type==1: # TT-based annotations format
vproc.videoProcessor(self._data_path, self._list_classes)
else:
print 'WARNING: wrong annotation type code, videos could not be processed'
vproc.deleteProcessData(self._data_path+'videos/')
def predictVideoSetDefault(videos_path, nmodel, save_path):
videosList = os.listdir(videos_path) # Lists all files (and directories) in the folder
for video in videosList:
if os.path.isfile(os.path.join(videos_path, video)):
labels = predictVideoDefault(video, videos_path, nmodel, save_path)
# # Clean all the remaining files created by OpenFace
vproc.deleteProcessData(videos_path)
def verifyVideoQuality(self, video_path, video_file):
os.system('./faceDetectorExtraction.sh '+video_file+' '+video_path)
video_name = os.path.splitext(video_file)[0]
videoPassesTest, cause = qver.verifyQuality(video_path, video_name)
if videoPassesTest!=True:
print 'The video passed the quality verification test'
else:
print 'The video did not pass the quality verification test -- reason: '+cause
vproc.deleteProcessData(video_path)
def predictVideoSetCustom(self, videos_path, model_name, save_path, input_type):
model_path = self._data_path+'models/saved/'
model_folder = model_name+'/'
videosList = os.listdir(videos_path) # Lists all files (and directories) in the folder
img_width = 90
if input_type == dataset.INPUT_FULLFACE:
img_height = 90
else:
img_height = 30
for video in videosList:
if os.path.isfile(os.path.join(videos_path, video)):
predictVideoGeneral(video, videos_path, model_name, save_path, model_path, model_folder, img_height, img_width)
# # Clean all the remaining files created by OpenFace
vproc.deleteProcessData(videos_path)
def verifyVideoSetQuality(self):
qver.runQualityVerification(self._data_path+'videos/',self._data_path+'discarded_videos.txt')
vproc.deleteProcessData(self._data_path+'videos/')