def getFace(videofile, tagfile, model='hog'):
print("tagfile: ", tagfile)
fileWriter = open(tagfile, "wt")
vh = cv_video_helper.VideoHelper()
for num_retrieved, frame in vh.frame_iterator(videofile, every_ms=1000):
print("num_retrieved: ", num_retrieved)
# Convert the image from BGR color (which OpenCV uses) to RGB color (which face_recognition uses)
rgb = frame[:, :, ::-1]
face_locations = face_recognition.face_locations(rgb, model=model)
height = rgb.shape[0]
width = rgb.shape[1]
fileWriter.write(str(num_retrieved))
if len(face_locations) > 0:
fileWriter.write(",")
first = True
for face_location in face_locations:
# Print the location of each face in this image
top, right, bottom, left = face_location
# x ,y, width, height
tem = '%s %s %s %s' % ('%.16f' %
(float(left) / width), '%.16f' %
(float(bottom) / height), '%.16f' %
(float(right - left) / width), '%.16f' %
(float(bottom - top) / height))
if first:
fileWriter.write(tem)
first = False
else:
fileWriter.write(" " + tem)
fileWriter.write("\n")
fileWriter.close()
def process(videofile, tfrecordDir=''):
file_arr = os.path.splitext(videofile)
if (tfrecordDir): # valid
filename = os.path.basename(videofile).split(".")[0]
output_tfrecords_file = "%s_output.tfrecord" % (tfrecordDir + '/' +
str(filename))
else:
output_tfrecords_file = "%s_output.tfrecord" % file_arr[0]
FLAGS.output_tfrecords_file = output_tfrecords_file
print("output_tfrecords_file", output_tfrecords_file)
extractor = feature_extractor.YouTube8MFeatureExtractor(FLAGS.model_dir)
writer = tf.python_io.TFRecordWriter(FLAGS.output_tfrecords_file)
total_written = 0
total_error = 0
labels = "0"
vh = cv_video_helper.VideoHelper()
for num_retrieved, rgb in vh.frame_iterator(videofile,
every_ms=1000.0 /
FLAGS.frames_per_second):
rgb_features = []
features = extractor.extract_rgb_frame_features(rgb[:, :, ::-1])
rgb_features.append(_bytes_feature(quantize(features)))
if not rgb_features:
printError('Could not get features for ' + videofile +
' ,frame order is ' + num_retrieved)
total_error += 1
continue
# Create SequenceExample proto and write to output.
feature_list = {
FLAGS.image_feature_key:
tf.train.FeatureList(feature=rgb_features),
}
if FLAGS.insert_zero_audio_features:
feature_list['audio'] = tf.train.FeatureList(
feature=[_bytes_feature(_make_bytes([0] * 128))] *
len(rgb_features))
# print(feature_list)
# E:/work/ai_script/tmp.MP4 -> E:/work/ai_script/tmp/%d.MP4
# url encode
virtualPath = quote(
file_arr[0]) + '/' + str(num_retrieved) + file_arr[1]
print("virtualPath = ", virtualPath)
example = tf.train.SequenceExample(
context=tf.train.Features(
feature={
FLAGS.labels_feature_key:
_int64_list_feature(sorted(map(int, labels.split(';')))),
FLAGS.video_file_key_feature_key:
_bytes_feature(_make_bytes(map(ord, virtualPath))),
}),
feature_lists=tf.train.FeatureLists(feature_list=feature_list))
writer.write(example.SerializeToString())
total_written += 1
writer.close()
print('Successfully encoded %i out of %i videos' %
(total_written, total_written + total_error))