import argparse
import utils
from FCECorpusHandler import FCECorpusHandler
""" Module to parse XML in Python using minidom
"""
if __name__ == '__main__':
parser = argparse.ArgumentParser(
description='Extracts transcript and summary from FCE Corpus.')
parser.add_argument('--fce_xml_dir',
type=str,
default=utils.project_dir_name() + '/data/',
help='FCE Corpus download directory')
parser.add_argument('--results_dir',
type=str,
default=utils.project_dir_name() + '/data/fce_txt/',
help='FCE Corpus txt format')
args = parser.parse_args()
fceCorpusHandler = FCECorpusHandler(args)
fceCorpusHandler.get_train_dev_test_sets()
fceCorpusHandler.xml_to_txt(data_type="train")
fceCorpusHandler.xml_to_txt(data_type="dev")
fceCorpusHandler.xml_to_txt(data_type="test")
import unittest
import utils
import os
import pandas as pd
from collections import defaultdict
"""
Handles .dat emotion information
"""
__author__ = "Gwena Cunha"
params = {'root': utils.project_dir_name() + 'data_test/BMI/', 'seconds': 10}
class DatEmotionHandler:
def __init__(self, root_dir, dat_filename='intended_1.dat'):
self.root_dir = root_dir
self.dat_filename = dat_filename
self.dat_file = utils.get_file(self.root_dir, self.dat_filename)
self.dat_sentences = self.dat_file.read().splitlines()
self.csv_data = None
def dat_to_csv(self, csv_filename='intended_1.csv'):
"""
Transform .dat files into .csv files with time, valence_score_raw, and valence_score emotion information
:param csv_filename: filename where dataframe will be saved to in csv format
:return:
"""
self.csv_data = defaultdict(list)
for i in range(0, len(self.dat_sentences)):
"""
Test steps: takes a movie file, resizes, downsamples, does RGB2HSV color transformation and saves data in .npz,
saves new video, loads that video for testing, loads .npz, converts from HSV2RGB, saves in restored video for
testing.
DONE: add emotion and text
"""
__author__ = "Gwena Cunha"
params = {
'fps': 10,
'root': utils.project_dir_name() + 'data_test/',
'new_size': 100,
'sr': 16000
}
def load_video(filename):
# Load videos (fps = 30)
clip = VideoFileClip(filename)
# Resize to 100 x 100
clip_resized = clip.resize(newsize=(params['new_size'], params['new_size']))
print("clip: {}, resized: {}".format(clip.size, clip_resized.size))
# Downsample
downsampled_frames, new_filename = utils.downsample_video(clip_resized, params, save_downsampled=True)
"""
Pre-processes data considering already spliced video and audio only
Synchronize video and audio
TODO: add emotion and text
"""
__author__ = "Gwena Cunha"
params = {
'fps': 10,
# 'root': '/media/ceslea/DATA/VideoEmotion/DataWithEmotionTags_noText_correctedAudio_hsio/',
'root': utils.project_dir_name() + 'data/cognimuse_10secs/',
'new_size': 100,
'sr': 16000,
'results_dir': utils.project_dir_name() + 'data/cognimuse_10secs/',
'seconds': 10,
'audio_len': 10*16000, # seconds*sr
}
def load_video(filename, params_substitute=None):
# To use in external scripts
if params_substitute is not None:
params = params_substitute
# Load videos (fps = 30)
clip = VideoFileClip(filename)
import utils
import glob
from natsort import natsorted
import pandas as pd
from scipy.io.wavfile import read
from splices2npz import load_video, process_audio
"""
Pre-processes data considering already spliced video and audio only
Synchronize video and audio
"""
__author__ = "Gwena Cunha"
is_test = False
if is_test:
ROOT = utils.project_dir_name() + 'data/deap/test_data2/'
else:
ROOT = utils.project_dir_name() + 'data/deap30frames/mp4/'
params = {
'fps': 10,
'root': ROOT,
'new_size': 100, # new frame size (100x100)
'sr': 16000,
'audio_len': 48000,
'results_dir': ROOT,
'seconds': 3,
}
def save_npz(videos,
type='train',
LOR: 75 (0: 57, 1: 18), 1239.75secs
30 seconds (2D emotion) BMI: 62 (0: 18, 1: 42, 2: 2, 3: 0)
CHI: 60 (0: 0, 1: 0, 2: 29, 3: 31)
CRA: 53 (0: 23, 1: 29, 2: 0, 3: 1)
DEP: 60 (0: 8, 1: 20, 2: 29, 3: 3)
FNE: 60 (0: 3, 1: 6, 2: 46, 3: 5)
GLA: 60 (0: 2, 1: 11, 2: 23, 3: 24)
LOR: 75 (0: 22, 1: 35, 2: 11, 3: 7)
"""
__author__ = "Gwena Cunha"
params = {
'root': utils.project_dir_name() + 'data/cognimuse_3secs/', #'data/cognimuse_multimodal_robust_bert/',
'sr': 16000,
'seconds': 3,
'num_samples': -1 # default = -1 for all samples. Use different to test code
}
def splice_video(video_clip, num_samples=-1, params_substitute=None):
# To use in external scripts
if params_substitute is not None:
params = params_substitute
# Load videos (fps = 30)
vid_fps = round(video_clip.fps)
vid_num_frames = round(vid_fps * video_clip.duration)
vid_frames_per_splice = params['seconds'] * vid_fps
print("Video - num frames: {}, size: {}, fps: {}, frames_per_splice: {}".
import argparse
import utils
from AMICorpusHandler import AMICorpusHandler
""" Module to parse XML in Python using minidom
"""
if __name__ == '__main__':
parser = argparse.ArgumentParser(
description='Extracts transcript and summary from AMI Corpus.')
parser.add_argument('--ami_xml_dir',
type=str,
default=utils.project_dir_name() + '/data/',
help='AMI Corpus download directory')
parser.add_argument('--results_transcripts_speaker_dir',
type=str,
default=utils.project_dir_name() +
'/data/ami-transcripts-speaker/',
help='AMI Corpus transcripts per speaker')
parser.add_argument('--results_transcripts_dir',
type=str,
default=utils.project_dir_name() +
'/data/ami-transcripts/',
help='AMI Corpus transcripts')
parser.add_argument('--results_summary_dir',
type=str,
default=utils.project_dir_name() +
'/data/ami-summary/',
help='AMI Corpus summaries')
args = parser.parse_args()
# print(args.ami_xml_dir)
Train (raw videos): same emotion as above, but number of splices per video is different:
Em 1 (total=2,069): Vid 1 (46), 2 (67), 3 (77), 4 (92), 5 (40), 6 (70), 7 (48), 9 (90), -> 530
10 (83), 11 (76), 12 (72), 13 (61), 15 (23), 20 (78), 22 (78), -> 471
23 (78), 24 (79), 25 (89), 27 (112), 29 (90), 30 (69), 31 (95), -> 612
32 (75), 34 (79), 35 (59), 36 (66), 37 (104), 38 (73) -> 456
Em 0 (total=781): Vid 8 (70), 14 (73), 16 (74), 19 (55), 21 (94), -> 366
26 (82), 28 (63), 33 (60), 39 (124), 40 (86) -> 415
P.S.: Raw 21, 23 have compromised frames and splicing doesn't work.
Solved by converting to .mov and back to .mp4
"""
__author__ = "Gwena Cunha"
is_test = False
if is_test:
ROOT = utils.project_dir_name() + 'data/deap/test_data/'
else:
ROOT = utils.project_dir_name() + 'data/deap_raw/mp4/'
params = {
'root': ROOT,
'emotion_root': utils.project_dir_name() + 'data/deap_raw/',
'sr': 16000,
'seconds': 3,
'num_samples':
-1 # default = -1 for all samples. Use different to test code
}
def get_num_video_splices(video_splices_dir):
return len(glob.glob(video_splices_dir + '*.mp4'))
tree = ET.parse(filename)
root = tree.getroot()
# one specific item attribute
print('Item #2 attribute:')
print(root[0][1].attrib)
# all item attributes
print('\nAll attributes:')
for elem in root:
for subelem in elem:
print(subelem.attrib)
# one specific item's data
print('\nItem #2 data:')
print(root[0][1].text)
# all items data
print('\nAll item data:')
for elem in root:
for subelem in elem:
print(subelem.text)
if __name__ == '__main__':
filename = utils.project_dir_name() + 'data/test.xml'
# test_minidom(filename)
test_element_tree(filename)
import pandas as pd
import csv
"""
DEAP has scores from 32 participants for 40 videos. This script will output a .csv file with the average emotion scores from all 32 participants.
Valence: 1 (most negative) - 9 (most positive) -> 5 is right in the middle
[1, 5) = neg
[5, 9] = pos
DEAP scale: valence, arousal (continuous 1-9) - Russell's scale ( J. A. Russell, “A circumplex model of affect,” Journal of Personality and Social Psychology, vol. 39, no. 6, pp. 1161–1178, 1980)
"Arousal can range from inactive (e.g. uninterested, bored) to active (e.g. alert, excited), whereas valence ranges from unpleasant (e.g. sad, stressed) to pleasant (e.g. happy, elated)"
Source: http://www.eecs.qmul.ac.uk/mmv/datasets/deap/doc/tac_special_issue_2011.pdf
"""
__author__ = "Gwena Cunha"
ROOT = utils.project_dir_name() + 'data/deap/'
def get_emotion_score_dictionary(filename='data/deap/participant_ratings.csv'):
with open(filename, newline='') as csvfile:
reader = csv.DictReader(csvfile)
# Fieldnames: Participant_id,Trial,Experiment_id,Start_time,Valence,Arousal,Dominance,Liking,Familiarity
valence_dict = defaultdict(lambda: [])
arousal_dict = defaultdict(lambda: [])
for row in reader:
# print(row['Trial'], row['Valence'], row['Arousal'])
valence_dict[row['Trial']].append(float(row['Valence']))
arousal_dict[row['Trial']].append(float(row['Arousal']))
dict = {'valence': valence_dict, 'arousal': arousal_dict}
return dict