test.py

import cv2
from facenet_pytorch import MTCNN

mtcnn = MTCNN(keep_all=True)


def resize_with_padding(image, target_size):
    target_width, target_height = target_size
    old_size = image.shape[:2]

    ratio = min(target_width / old_size[1], target_height / old_size[0])
    new_size = (int(old_size[1] * ratio), int(old_size[0] * ratio))

    resized_image = cv2.resize(image, new_size)

    delta_w = target_width - new_size[0]
    delta_h = target_height - new_size[1]
    top, bottom = delta_h // 2, delta_h - (delta_h // 2)
    left, right = delta_w // 2, delta_w - (delta_w // 2)

    color = [0, 0, 0]
    new_image = cv2.copyMakeBorder(resized_image, top, bottom, left, right, cv2.BORDER_CONSTANT, value=color)
    
    return new_image
def load_Video(num_frames=16):
    vid = cv2.VideoCapture('DeepFakeDetect\Celeb-DF-v2\Train\Real\Celeb_real_id11_0004.mp4')
    frame_cnt = vid.get(cv2.CAP_PROP_FRAME_COUNT)
    interval = int(max(1, frame_cnt//num_frames))
    cnt = 0
    print(cnt)
    currFrame = 0
    width = 0
    height = 0
    writer = cv2.VideoWriter('DeepFakeDetect\test.mp4',cv2.VideoWriter_fourcc(*"mp4v"),30,(224,224))
    while cnt <= num_frames and vid.isOpened():
        ret,frame = vid.read()
        if not ret:
            break
        if currFrame%interval == 0:
            boxes,_ = mtcnn.detect(frame)
            if boxes is not None:
                if cnt == 0:
                    width = boxes[0][3]-boxes[0][1]+10
                    height = boxes[0][2]-boxes[0][0]+10
                centerX = (boxes[0][1]+boxes[0][3])//2
                centerY = (boxes[0][0]+boxes[0][2])//2
                frame_cropped = frame[max(int(centerX-(width//2)),0):min(int(centerX+(width//2)),frame.shape[0]),max(int(centerY-(height//2)),0):min(int(centerY+(height//2)),frame.shape[1])]
                frame_cropped = resize_with_padding(frame_cropped,(224,224))
                writer.write(frame_cropped)
                cnt+=1
        currFrame+=1
    vid.release()
print('Done')

load_Video()