face_swap.py

import insightface
import os
import onnxruntime
import onnx
import cv2
import threading
import cv2
import numpy
import numpy as np
from deepface import DeepFace
import matplotlib.pyplot as plt
from gfpgan.utils import GFPGANer
import threading

available = onnxruntime.get_available_providers()

provider = [available[0]]

# In case iof GPU Error use ->
# provider = 'CPUExecutionProvider' 

print("Using", provider)

reference_face_position = 0
similar_face_distance = 0.85

swapper = insightface.model_zoo.get_model("models/inswapper_128.onnx", providers=provider)
enhancer = GFPGANer(model_path="models/GFPGANv1.4.pth", upscale=1, device="cpu")
lock = threading.Lock()
face_annalyser = None

def annalyser():
    global face_annalyser
    with lock:
        if face_annalyser is None:
            face_annalyser = insightface.app.FaceAnalysis(name='buffalo_l', providers=provider)
            face_annalyser.prepare(ctx_id=0)
    return face_annalyser

def get_many_faces(frame):
    try:
        return annalyser().get(frame)
    except ValueError:
        return None

def get_one_face(frame, position = 0):
    many_faces = get_many_faces(frame)
    if many_faces:
        try:
            return many_faces[position]
        except IndexError:
            return many_faces[-1]
    return None

def find_similar_face(frame, reference_face):
    many_faces = get_many_faces(frame)
    if many_faces:
        for face in many_faces:
            if hasattr(face, 'normed_embedding') and hasattr(reference_face, 'normed_embedding'):
                distance = numpy.sum(numpy.square(face.normed_embedding - reference_face.normed_embedding))
                if distance < similar_face_distance:
                    return face
    return None

def enhance_face(target_face, temp_frame):
    start_x, start_y, end_x, end_y = map(int, target_face['bbox'])
    padding_x = int((end_x - start_x) * 0.5)
    padding_y = int((end_y - start_y) * 0.5)
    start_x = max(0, start_x - padding_x)
    start_y = max(0, start_y - padding_y)
    end_x = max(0, end_x + padding_x)
    end_y = max(0, end_y + padding_y)
    temp_face = temp_frame[start_y:end_y, start_x:end_x]
    if temp_face.size:
        with threading.Semaphore():
            _, _, temp_face = enhancer.enhance(
                temp_face,
                paste_back=True
            )
        temp_frame[start_y:end_y, start_x:end_x] = temp_face
    return temp_frame

def process_frame(source_face, reference_face, temp_frame):
    target_face = find_similar_face(temp_frame, reference_face)
    if target_face:
        try:
            temp_frame = swapper.get(temp_frame, target_face, source_face, paste_back=True)
            temp_frame = enhance_face(source_face, temp_frame)
        except:
            print("Face not found in the image provided")
    return temp_frame


def swap(new_face,image): 
    source_face = get_one_face(new_face)
    target_frame = image
    reference_face = get_one_face(target_frame, reference_face_position)
    result = process_frame(source_face, reference_face, target_frame)
    return result

def face_image_resize(img, w,h):
    detected_faces = []
    try:
        detected_faces = DeepFace.analyze(img, actions = ["gender"], enforce_detection=False)
    except: 
        print("No face detected")
    detected_face = img
    if len(detected_faces) > 0:
        for temp_img in detected_faces[:1]:
            x = max(temp_img["region"]["x"] - int(temp_img["region"]["w"]/2),0)
            y = max(temp_img["region"]["y"] - int(temp_img["region"]["h"]/2),0)
            w = temp_img["region"]["w"] + int(temp_img["region"]["w"]/1.5) 
            h = temp_img["region"]["h"] + int(temp_img["region"]["h"]/1.5) 
            detected_face = img[y:y+h, x:x+w]
    new_image = cv2.resize(detected_face, (w, h))
    return new_image

def replace_face(path, images=[]):
    img = cv2.imread(path)
    obj = []
    male = 0
    female = 0
    try:
        obj = DeepFace.analyze(img, actions = ["gender"])
    except:
        print("No face detected in image provided")
    for temp_img in obj:
        x = max(temp_img["region"]["x"] - int(temp_img["region"]["w"]/2),0)    
        y = max(temp_img["region"]["y"] - int(temp_img["region"]["h"]/2),0)
        w = temp_img["region"]["w"] + int(temp_img["region"]["w"]/1.5) 
        h = temp_img["region"]["h"] + int(temp_img["region"]["h"]/1.5)           
        plt.imshow(img[temp_img["region"]["y"]:temp_img["region"]["y"]+temp_img["region"]["h"], temp_img["region"]["x"]:temp_img["region"]["x"]+temp_img["region"]["w"]][:,:,::-1])
        plt.show()
        if "yes" in input("Change face ? yes/no : ").lower():
            if len(images) > 0:
                img[y:y+h, x:x+w] = swap(cv2.imread(images[0]), img[y:y+h, x:x+w])
                del images[-1]
            else:
                try:
                    if "woman" in temp_img["dominant_gender"].lower():
                        img[y:y+h, x:x+w] = swap(cv2.imread(f'data/female/{female}.jpg'), img[y:y+h, x:x+w])
                        female = female + 1
                    else:
                        img[y:y+h, x:x+w] = swap(cv2.imread(f'data/male/{male}.jpg'), img[y:y+h, x:x+w])
                        male = male + 1
                except:
                    print("Out of new face images for this gender")
    cv2.imwrite("output/output.jpg", img)