app.py

import gradio as gr

import cv2
from mtcnn.mtcnn import MTCNN
import tensorflow as tf
import tensorflow_addons
import numpy as np


detector = MTCNN()

model = tf.keras.models.load_model("FINAL-EFFICIENTNETV2-B0")


def deepfakespredict(input_img):

    face = detector.detect_faces(input_img)

    text =""

    if len(face) > 0:
        x, y, width, height = face[0]['box']
        x2, y2 = x + width, y + height

        cv2.rectangle(input_img, (x, y), (x2, y2), (0, 255, 0), 2)

        face_image = input_img[y:y2, x:x2]
        face_image2 = cv2.cvtColor(face_image, cv2.COLOR_BGR2RGB)
        face_image3 = cv2.resize(face_image2, (224, 224))
        face_image4 = face_image3/255

        pred = model.predict(np.expand_dims(face_image4, axis=0))[0]

        if pred[1] >= 0.6:
            text = "The image is fake."
        elif pred[0] >= 0.6:
            text = "The image is real."
        else:
            text = "The image might be real or fake."

        # if pred[1] >= 0.5:
        #     text = "The image is fake."
        # else:
        #     text = "The image is real."

    else:
        text = "Face is not detected in the image."

    return pred, text, input_img


title="EfficientNetV2 Deepfakes Image Detector"
description="This is a demo implementation of EfficientNetV2 Deepfakes Image Detector. To use it, simply upload your image, or click one of the examples to load them."
examples = [
            ['fake-86.jpg'],
            ['fake-239.jpg'],
            ['fake-254.jpg'],
            ['fake-1266.jpg'],
            ['fake-2225.jpg']
            ]
demo = gr.Interface(deepfakespredict,
                     inputs = ["image"],
                     outputs=["text","text","image"],
                     title=title,
                     description=description,
                     examples=examples
                     )
demo.launch()