From d3920a26a25bbfb93ac81bf6e3020f552e123606 Mon Sep 17 00:00:00 2001 From: Vidya Koesmahargyo Date: Wed, 25 Nov 2020 09:59:42 -0500 Subject: [PATCH] facial tremor testing --- dbm_lib/controller/process_feature.py | 6 +- .../raw_features/movement/__init__.py | 1 + .../raw_features/movement/facial_tremor.py | 159 ++++++++++++++++++ .../raw_features/util/math_util.py | 57 +++++++ resources/features/facial/config.json | 1 + 5 files changed, 223 insertions(+), 1 deletion(-) create mode 100644 dbm_lib/dbm_features/raw_features/movement/facial_tremor.py create mode 100644 dbm_lib/dbm_features/raw_features/util/math_util.py create mode 100644 resources/features/facial/config.json diff --git a/dbm_lib/controller/process_feature.py b/dbm_lib/controller/process_feature.py index f5112da2..488b835c 100644 --- a/dbm_lib/controller/process_feature.py +++ b/dbm_lib/controller/process_feature.py @@ -7,7 +7,7 @@ created: 2020-20-07 from dbm_lib.dbm_features.raw_features.audio import intensity, pitch_freq, hnr, gne, voice_frame_score, formant_freq from dbm_lib.dbm_features.raw_features.audio import pause_segment, jitter, shimmer, mfcc from dbm_lib.dbm_features.raw_features.video import face_asymmetry, face_au, face_emotion_expressivity, face_landmark -from dbm_lib.dbm_features.raw_features.movement import head_motion, eye_blink, voice_tremor +from dbm_lib.dbm_features.raw_features.movement import head_motion, eye_blink, voice_tremor, facial_tremor import subprocess import logging @@ -120,6 +120,7 @@ def process_movement(video_uri, out_dir, dbm_group, r_config, dlib_model): return logger.info('Processing movement variables from data in {}'.format(video_uri)) + logger.info('processing head movement....') head_motion.run_head_movement(video_uri, out_dir, r_config) @@ -129,6 +130,9 @@ def process_movement(video_uri, out_dir, dbm_group, r_config, dlib_model): logger.info('processing voice tremor....') voice_tremor.run_vtremor(video_uri, out_dir, r_config) + logger.info('processing facial tremor....') + face_tremor.fac_tremor_process(video_uri, out_dir, r_config, model_output=True) + def remove_file(file_path): """ removing wav file diff --git a/dbm_lib/dbm_features/raw_features/movement/__init__.py b/dbm_lib/dbm_features/raw_features/movement/__init__.py index dca2f624..03fd8909 100644 --- a/dbm_lib/dbm_features/raw_features/movement/__init__.py +++ b/dbm_lib/dbm_features/raw_features/movement/__init__.py @@ -13,3 +13,4 @@ import os DBMLIB_PATH = os.path.dirname(__file__) DBMLIB_VTREMOR_LIB = os.path.abspath(os.path.join(DBMLIB_PATH, '../../../../resources/libraries/voice_tremor.praat')) +DBMLIB_FTREMOR_CONFIG = os.path.abspath(os.path.join(DBMLIB_PATH, '../resources/features/facial/config.json')) diff --git a/dbm_lib/dbm_features/raw_features/movement/facial_tremor.py b/dbm_lib/dbm_features/raw_features/movement/facial_tremor.py new file mode 100644 index 00000000..e6bfb1a5 --- /dev/null +++ b/dbm_lib/dbm_features/raw_features/movement/facial_tremor.py @@ -0,0 +1,159 @@ +import sys, os, glob, cv2, re +import pickle, json +import pandas as pd +import numpy as np +import numpy.ma as ma +import logging +from os.path import join + +from dbm_lib.dbm_features.raw_features.util import util as ut +from dbm_lib.dbm_features.raw_features.util.math_util import * + +from dbm_lib.dbm_features.raw_features.movement import DBMLIB_FTREMOR_CONFIG + +logging.basicConfig(level=logging.INFO) +logger=logging.getLogger() + +ft_dir = 'movement/facial_tremor' +csv_ext = '_fac_tremor.csv' +model_ext = '_fac_model.csv' +fac_features_ext = '_fac_features.csv' + +def compute_features(out_dir, df_of, r_config): + """ Computes features + + Returns: features in vector format + """ + config = json.loads(open(DBMLIB_FTREMOR_CONFIG,'r').read()) + logger.info('json file read') + pattern_x = re.compile("l\d+_x") + pattern_y = re.compile("l\d+_y") + + # assumption: distance of face to camera remains at roughly static + + # logic break + landmark_columns = [] + for col in df_of.columns: + if pattern_x.match(col) or pattern_y.match(col): + landmark_columns.append(col) + + df_of= df_of[(df_of[landmark_columns]!= 0).any(axis=1)] + df_of.reset_index(inplace=True) + + num_frames = len(df) + logger.info("Number of frames to be processed: {}".format(str(num_frames))) + landmarks = config['landmarks'] + + try: + if num_frames == 0: + error_reason = "No frames with visible face." + logger.error(error_reason) + return empty_frame(landmarks, r_config, error_reason) + +# if num_frames < 60: +# error_reason = 'Number of frames with visible face < 60. Video too short' +# logger.error(error_reason) +# return empty_frame(landmarks, f_cfg, error_reason) + + first_row = df_of.iloc[0] + + facew = abs(first_row[config['face_width_left']] - first_row[config['face_width_right']]) + faceh = abs(first_row[config['face_height_left']] - first_row[config['face_height_right']]) + + if facew == 0 or faceh == 0: + error_reason = 'face width or height = 0. Check landmark values' + logger.error(error_reason) + return empty_frame(landmarks, r_config) + + fac_disp = calc_displacement_vec(df_of, landmarks, num_frames) + + # if verbose: + # logger.info("Displacement output: {}".format(str(fac_disp))) + + fac_disp_median = np.median(fac_disp, axis = 1) + fac_disp_mean = np.mean(fac_disp, axis = 1) + + if len(fac_disp.shape)!=2: + error_reason = 'fac_disp is not 2D. smth went wrong with disp calc' + logger.error(error_reason) + return empty_frame(landmarks, r_config, error_reason) + + if len(fac_disp[0])<=1: + error_reason = 'Video too short. smth went wrong with disp calc' + logger.error(error_reason) + return empty_frame(landmarks, r_config, error_reason) + + fac_corr_mat = np.corrcoef(fac_disp, rowvar = True) + # extract relevant row from cov matrix + ref_lmk_index = [i for i, lmk in enumerate(landmarks) if config['ref_lmk']==lmk] + fac_corr = fac_corr_mat[ref_lmk_index][0] + + fac_area = config['ref_area'] / (facew * faceh) + + # if verbose: + # logger.info("Face area: {}".format(fac_area)) + # logger.info("Face Displacement Median: {}".format(str(fac_disp_median))) + # logger.info("Face Displacement Mean: {}".format(str(fac_disp_mean))) + + fac_features1 = np.multiply(fac_area * fac_disp_median, (1. - fac_corr)) + fac_features2 = np.multiply(fac_area * fac_disp_mean, (1. - fac_corr)) + +# base_fac_features = np.dot(fac_area * fac_disp_median, (1. - fac_corr)) + + fac_features_dict = {} + for i, landmark in enumerate(landmarks): + fac_features_dict['fac_features_mean_{}'.format(landmark)] = [fac_features2[i]] + raw_variable_map = 'fac_tremor_median_{}'.format(landmark) + fac_features_dict[r_config.raw_feature[raw_variable_map]] = [fac_features1[i]] + + fac_features_dict['fac_disp_median_{}'.format(landmark)] = [fac_disp_median[i]] + fac_features_dict['fac_corr_{}'.format(landmark)] = [fac_corr[i]] + + fac_features_dict[r_config.err_reason] = [''] + data = pd.DataFrame.from_dict(fac_features_dict) + logger.info('Concluded computing tremor features') + + return data + + except Exception as e: + logger.error('Error computing tremor features: {}'.format(str(e))) + return empty_frame(landmarks, r_config, str(e)) + +def empty_frame(landmarks, r_config, error_reason): + fac_features_dict = {} + for i, landmark in enumerate(landmarks): + raw_variable_map = 'fac_tremor_median_{}'.format(landmark) + fac_features_dict[r_config.raw_feature[raw_variable_map]] = [np.nan] + + fac_features_dict['fac_features_mean_{}'.format(landmark)] = [np.nan] + fac_features_dict['fac_disp_median_{}'.format(landmark)] = [np.nan] + fac_features_dict['fac_corr_{}'.format(landmark)] = [np.nan] + + fac_features_dict[r_config.err_reason] = [error_reason] + empty_frame = pd.DataFrame.from_dict(fac_features_dict) + return empty_frame + +def fac_tremor_process(video_uri,out_dir,r_config, model_output=False): + """ + processing input videos + """ + try: + logger.info('filtering path: ',video_uri,out_dir) + input_loc, out_loc, fl_name = ut.filter_path(video_uri, out_dir) + of_csv_path = glob.glob(join(out_loc, fl_name + '_OF_features/*.csv')) + + if len(of_csv_path)>0: + of_csv = of_csv_path[0] + df_of = pd.read_csv(of_csv, error_bad_lines=False) + + logger.info('Processing Output file {} '.format(os.path.join(out_loc, fl_name))) + + feats = compute_features(of_csv_path , df_of, r_config) + if model_output: + result = score(feats, r_config) + feats = pd.concat([feats, result], axis=1) + + ut.output_audio_feature(feats, new_out_base_dir, '/'+fac_dir, fac_ext) + + except Exception as e: + logger.error('Failed to process video file') diff --git a/dbm_lib/dbm_features/raw_features/util/math_util.py b/dbm_lib/dbm_features/raw_features/util/math_util.py new file mode 100644 index 00000000..283acadc --- /dev/null +++ b/dbm_lib/dbm_features/raw_features/util/math_util.py @@ -0,0 +1,57 @@ +""" +file_name: facial_tremor +project_name: cdx_analysis +created: 2019-03-16 +author: Deshana Desai +""" +import sys, os, glob, cv2 +import pandas as pd +import numpy as np + + +def euclidean_distance(point1, point2): + """ + Compute euclidean distance between points + """ + + return np.sqrt((point1[0] - point2[0])**2 + (point1[1] - point2[1])**2) + + +# def detect_peaks() + + +def expand_landmarks(landmarks): + """ + util method to expand landmark list: + eg: [1,2] -> [['l1_x', 'l1_y'], ['l2_x', 'l2_y']] + """ + return [['l{}_x'.format(l), 'l{}_y'.format(l)] for l in landmarks] + + + +def calc_displacement_vec(df, landmarks, num_frames): + """ + Calculates displacement vector frame by frame + """ + + landmarks = expand_landmarks(landmarks) + + disp_vec = np.zeros((len(landmarks), num_frames)) + prev_point = np.zeros((len(landmarks), 2)) + + # initialize + for j, pair in enumerate(landmarks): + first_row = df.iloc[0] + prev_point[j] = (first_row[pair[0]], first_row[pair[1]]) + + + for i in range(num_frames): + frame_row = df.iloc[i] + for j, pair in enumerate(landmarks): + x, y = pair[0], pair[1] + current = (frame_row[x], frame_row[y]) + deviation = euclidean_distance( current, prev_point[j]) + disp_vec[j][i] = deviation + prev_point[j] = current + + return disp_vec diff --git a/resources/features/facial/config.json b/resources/features/facial/config.json new file mode 100644 index 00000000..ebad1b06 --- /dev/null +++ b/resources/features/facial/config.json @@ -0,0 +1 @@ +{"ref_lmk": 28, "ref_area": 350000, "face_width_left": "l15_x", "face_width_right": "l1_x", "face_height_left": "l8_y", "face_height_right": "l27_y", "landmarks": [5, 12, 8, 48, 54, 28, 51, 66, 57], "model_path": "resources/facial/svm_bin_fac_tremor.sav", "feature_order": ["fac_features_mean_5", "fac_features_mean_12", "fac_features_mean_8", "fac_features_mean_48", "fac_features_mean_54", "fac_features_mean_28", "fac_features_mean_51", "fac_features_mean_66", "fac_features_mean_57", "fac_features_median_5", "fac_features_median_12", "fac_features_median_8", "fac_features_median_48", "fac_features_median_54", "fac_features_median_28", "fac_features_median_51", "fac_features_median_66", "fac_features_median_57"]}