visualize/visualize.py

import sys
import pickle
import cv2
import numpy as np
from moviepy.editor import VideoFileClip
from moviepy.audio.io.AudioFileClip import AudioFileClip


def get_label_str(value_list):
    labels = ['background', 'wearer speaking start', 'wearer speaking', 'other person speaking']
    decoded_labels = []
    for label_str in value_list:
        binary_string = ''.join(map(lambda x: '1' if x else '0', label_str))
        temp = []
        cnt = 0
        for i, bit in enumerate(binary_string):
            if bit == '1':
                cnt+=1
                temp.append(labels[i])
        if cnt > 1:
            temp = ' & '.join(temp)
            print(binary_string, temp)
        else:
            temp = temp[0]
        decoded_labels.append(temp)

    return decoded_labels


def view_pkl(pickle_file):
    with open(pickle_file, 'rb') as f:
        data = pickle.load(f)
        predict_list = data['prediction']
        target_list = data['target']
    target_value_list = get_label_str(target_list)
    new_list = []
    max_list = []
    for row in predict_list:
        new_row = [round(value / 10000, 3) for value in row]
        new_list.append(new_row)
        max_index = np.argmax(row)
        result = np.zeros_like(row)
        result[max_index] = 1
        max_list.append(result)
    max_value_list = get_label_str(max_list)
    predict_list = new_list

    return predict_list, target_value_list, max_value_list, target_list


def draw_text(img, text,
          font=cv2.FONT_HERSHEY_SIMPLEX,
          pos=(0, 0),
          font_scale=1,
          font_thickness=2,
          text_color=(255, 255, 255),
          text_color_bg=(0, 0, 0)
          ):

    x, y = pos
    text_size, _ = cv2.getTextSize(text, font, font_scale, font_thickness)
    text_w, text_h = text_size
    cv2.rectangle(img, pos, (x + text_w, y + text_h), text_color_bg, -1)
    cv2.putText(img, text, (x, y + text_h + font_scale - 1), font, font_scale, text_color, font_thickness)

    return text_size


def video_anno(predict_list, target_list, max_list, input_file_path, output_file_path, test_list):
    cap = cv2.VideoCapture(input_file_path)
    filename = input_file_path.split('/')[1].split('.')[0]

    frame_width = int(cap.get(3))
    frame_height = int(cap.get(4))
    fps = int(cap.get(5))
    frame_count = int(cap.get(7))

    fourcc = cv2.VideoWriter_fourcc(*'mp4v')
    out = cv2.VideoWriter(output_file_path, fourcc, fps, (frame_width, frame_height))

    print(len(predict_list))
    frame_number = 0
    anno_num = 0
    sig_num = 0
    while cap.isOpened():
        ret, frame = cap.read()
        if not ret:
            break
        frame_number+=1
        sig_num+=1
        if anno_num < len(predict_list):
            #print(anno_num)
            if frame_number>=384 and sig_num==6:
                sig_num = 0
                text_GT = f'Ground Truth: {frame_number} {target_list[anno_num]}'
                text_PR = f'Model Predict: {frame_number} {max_list[anno_num]}'
                text_DT = f'Predict detail: {frame_number} {predict_list[anno_num]}'
                anno_num+=1
            if sig_num > 5:
                sig_num = 0

        if frame_number >= 384:
            draw_text(frame, text_GT, pos=(10, 10))
            draw_text(frame, text_PR, pos=(10, 50))
            draw_text(frame, text_DT, pos=(10, 100))

        out.write(frame)

    cap.release()
    out.release()
    cv2.destroyAllWindows()


def main(save, save2, pickle_file, audio_path):
    video = '/original/video/path.mp4'

    video_clip = VideoFileClip(video)
    audio_clip = video_clip.audio
    audio_clip.write_audiofile(audio_path)

    predict_list, target_list, max_list, test_list = view_pkl(pickle_file)
    video_anno(predict_list, target_list, max_list, video, save, test_list) # save => 'test_video/00792fa8-988c-4c85-8e80-73eb3ac53e80_anno2.mp4'

    video_clip = VideoFileClip(save)
    audio_clip = AudioFileClip(audio_path)
    video_clip = video_clip.set_audio(audio_clip)
    video_clip.write_videofile(save2, codec='libx264', audio_codec='aac')

    video_clip.close()
    audio_clip.close()


if __name__ == '__main__':
    main(sys.argv[1], sys.argv[2], sys.argv[3], sys.argv[4])