Code for imaginaryCTF forensics "Subtitles" challenge

subtitles_solve.py

import ffmpeg
from tensorflow import keras
import cv2


def video_to_data():
    # Extract frames from video every 1 second and save to data dir using ffmpeg and start from 0
    ffmpeg.input("./subtitles.mp4").output("./data/%d.png", ss="0").run()


def main():
    labels = open("./label.txt", "rt").read().split("\n")
    video_to_data()
    # Load model.h5 to keras using GPU
    model = keras.models.load_model("./model.h5")
    res = []

    # Loop every images in data dir and compare label with prediction
    for i in range(len(labels) - 1):
        image = cv2.imread(f"./data/{i + 1}.png")[:, :, 0]
        label = model.predict(image.reshape(1, 28, 28, 1))
        label = label.argmax()
        if str(label) != labels[i]:
            res.append("1")
        else:
            res.append("0")

    print("".join(res))


if __name__ == "__main__":
    main()

awk 'NR % 4 == 0' subtitles.srt > label.txt

https://drive.google.com/file/d/1mwtvDnndni3c4sG8Q-HdQkfFRrUxJ_oo/view?usp=sharing