【部分原创】python实现视频内的face swap(换脸)

1.准备工作，按博主的环境为准

Python 3.5

Opencv 3

Tensorflow 1.3.1

Keras 2

cudnn和CUDA，如果你的GPU足够厉害并且支持的话，可以选择安装

那就先安装起来，有兴趣的朋友给我个暗示，好让我有动力写下去，想实现整套的功能还是有点复杂的

第一部分，数据采集，及视频内人物脸

import cv2

save_path = 'your save path'
cascade = cv2.CascadeClassifier('haarcascade_frontalface_alt.xml path')
cap = cv2.VideoCapture('your video path')
i = 0
while True:
    ret,frame = cap.read()
    gray = cv2.cvtColor(frame,cv2.COLOR_BGR2GRAY)
    rect = cascade.detectMultiScale(gray,scaleFactor=1.3,minNeighbors=9,minSize=(50,50),flags = cv2.CASCADE_SCALE_IMAGE)
    print ("rect",rect)
    if not rect is ():
        for x,y,z,w in rect:
            roiImg = frame[y:y+w,x:x+z]
            cv2.imwrite(save_path+str(i)+'.jpg',roiImg)
            cv2.rectangle(frame,(x,y),(x+z,y+w),(0,0,255),2)
            i +=1
    cv2.imshow('frame',frame)
    if cv2.waitKey(1) &0xFF == ord('q'):
        break
cap.release()
cv2.destroyAllWindows()

 

第二部分，国外大神开源代码，用于模型训练

import cv2
import numpy

from utils import get_image_paths, load_images, stack_images
from training_data import get_training_data

from model import autoencoder_A
from model import autoencoder_B
from model import encoder, decoder_A, decoder_B

try:
    encoder  .load_weights( "models/encoder.h5"   )
    decoder_A.load_weights( "models/decoder_A.h5" )
    decoder_B.load_weights( "models/decoder_B.h5" )
except:
    pass

def save_model_weights():
    encoder  .save_weights( "models/encoder.h5"   )
    decoder_A.save_weights( "models/decoder_A.h5" )
    decoder_B.save_weights( "models/decoder_B.h5" )
    print( "save model weights" )

images_A = get_image_paths( "data/trump" )
images_B = get_image_paths( "data/cage"  )
images_A = load_images( images_A ) / 255.0
images_B = load_images( images_B ) / 255.0

images_A += images_B.mean( axis=(0,1,2) ) - images_A.mean( axis=(0,1,2) )

print( "press 'q' to stop training and save model" )

for epoch in range(1000000):
    batch_size = 64
    warped_A, target_A = get_training_data( images_A, batch_size )
    warped_B, target_B = get_training_data( images_B, batch_size )

    loss_A = autoencoder_A.train_on_batch( warped_A, target_A )
    loss_B = autoencoder_B.train_on_batch( warped_B, target_B )
    print( loss_A, loss_B )

    if epoch % 100 == 0:
        save_model_weights()
        test_A = target_A[0:14]
        test_B = target_B[0:14]

    figure_A = numpy.stack([
        test_A,
        autoencoder_A.predict( test_A ),
        autoencoder_B.predict( test_A ),
        ], axis=1 )
    figure_B = numpy.stack([
        test_B,
        autoencoder_B.predict( test_B ),
        autoencoder_A.predict( test_B ),
        ], axis=1 )

    figure = numpy.concatenate( [ figure_A, figure_B ], axis=0 )
    figure = figure.reshape( (4,7) + figure.shape[1:] )
    figure = stack_images( figure )

    figure = numpy.clip( figure * 255, 0, 255 ).astype('uint8')

    cv2.imshow( "", figure )
    key = cv2.waitKey(1)
    if key == ord('q'):
        save_model_weights()
        exit()

 

第三部分，国外大神开源代码，人脸输出

import cv2
import numpy
from pathlib import Path

from utils import get_image_paths

from model import autoencoder_A
from model import autoencoder_B
from model import encoder, decoder_A, decoder_B

encoder  .load_weights( "models/encoder.h5"   )
decoder_A.load_weights( "models/decoder_A.h5" )
decoder_B.load_weights( "models/decoder_B.h5" )

images_A = get_image_paths( "data/trump" )
images_B = get_image_paths( "data/cage" )

def convert_one_image( autoencoder, image ):
    assert image.shape == (256,256,3)
    crop = slice(48,208)
    face = image[crop,crop]
    face = cv2.resize( face, (64,64) )
    face = numpy.expand_dims( face, 0 )
    new_face = autoencoder.predict( face / 255.0 )[0]
    new_face = numpy.clip( new_face * 255, 0, 255 ).astype( image.dtype )
    new_face = cv2.resize( new_face, (160,160) )
    new_image = image.copy()
    new_image[crop,crop] = new_face
    return new_image

output_dir = Path( 'output' )
output_dir.mkdir( parents=True, exist_ok=True )

for fn in images_A:
    image = cv2.imread(fn)
    new_image = convert_one_image( autoencoder_B, image )
    output_file = output_dir / Path(fn).name
    cv2.imwrite( str(output_file), new_image )

 

第四部分，人脸替换

#import necessary libraries
import cv2
import glob as gb
# import numpy

#capture video from the webcam
cap = cv2.VideoCapture('your video path')
fourcc = cv2.VideoWriter_fourcc(*'XVID')
out = cv2.VideoWriter('your output video path', fourcc, 20.0, (1920, 1080))
#load the face finder
face_cascade = cv2.CascadeClassifier('haarcascade_frontalface_alt.xml path')

#load the face that will be swapped in
img_path = gb.glob("your image path")

#start loop
for path in img_path:
    face_img = cv2.imread(path)
    while True:
        ret, img = cap.read()  # read image
        gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
        faces = face_cascade.detectMultiScale(gray, 1.3, 3)  # find faces
        # for all the faces found in the frame
        for (x, y, w, h) in faces:
            # resize and blend the face to be swapped in
            face = cv2.resize(face_img, (h, w), interpolation=cv2.INTER_CUBIC)
            face = cv2.addWeighted(img[y:y + h, x:x + w], .5, face, .5, 1)
            # swap faces
            img[y:y + h, x:x + w] = face
            out.write(img)
            # show the image
        cv2.imshow('img', img)
        key = cv2.waitKey(1)
        if key == ord('q'):
            exit()
cap.release()
cv2.destroyAllWindows()

 

最后放一张训练一小时后的视频截图，用的是尼古拉斯凯奇的脸

面朝大海--\(˙<>˙)/--落霞与孤鹜齐飞