app.py

import os
import subprocess

os.system("pip install gradio==3.50")
os.system("pip install dlib==19.24.2")
os.system("pip install scikit-learn")
os.system("pip install scikit-image")
os.system("pip install tensorflow==2.11.0")


#############################################

import torch
print(f"Is CUDA available: {torch.cuda.is_available()}")
# True
print(f"CUDA device: {torch.cuda.get_device_name(torch.cuda.current_device())}")

###################################################

import tensorflow as tf
gpus = tf.config.list_physical_devices('GPU')
print("Available GPUs TF:", gpus)

if gpus:
    try:
        # Allow TensorFlow to allocate memory as needed
        for gpu in gpus:
            tf.config.experimental.set_memory_growth(gpu, True)
    except RuntimeError as e:
        print(e)
else:
    print("No GPUs available.")

###################################################


from skimage import exposure
from skimage.filters import laplace
from argparse import Namespace
import pprint
import numpy as np
from PIL import Image
import torch
import torchvision.transforms as transforms
import cv2
import dlib
import matplotlib.pyplot as plt
import gradio as gr  # Importing Gradio as gr
import tensorflow as tf
from tensorflow.keras.models import load_model
from tensorflow.keras.losses import MeanSquaredError
from tensorflow.keras.preprocessing.image import img_to_array
from huggingface_hub import hf_hub_download, login
from datasets.augmentations import AgeTransformer
from utils.common import tensor2im
from models.psp import pSp

# Huggingface login
login(token=os.getenv("TOKENKEY"))

########################################################################
############## tensorflow model for age calculation #######################

# If 'mse' is a custom function needed, 
#custom_objects = {'mse': MeanSquaredError()}
new_age_model = load_model("age_prediction_modelV2.h5")
########################################################################


########################################################################
############## pytorch model for age calculation #######################

#age_calc_model = torch.load('Custom_Age_prediction_model.pth')

########################################################################

# Download models from Huggingface
#age_prototxt = hf_hub_download(repo_id="AshanGimhana/Age_Detection_caffe", filename="age.prototxt")
#caffe_model = hf_hub_download(repo_id="AshanGimhana/Age_Detection_caffe", filename="dex_imdb_wiki.caffemodel")


########################################################################
############## caffe model for age calculation #######################

#age_prototxt = hf_hub_download(repo_id="AshanGimhana/Age_Detection_caffe", filename="age.prototxt")
#caffe_model = hf_hub_download(repo_id="AshanGimhana/Age_Detection_caffe", filename="dex_imdb_wiki.caffemodel")
# Age prediction model setup
#age_net = cv2.dnn.readNetFromCaffe(age_prototxt, caffe_model)
########################################################################


########################################################################################################
# Aging model #
sam_ffhq_aging = hf_hub_download(repo_id="AshanGimhana/Face_Agin_model", filename="sam_ffhq_aging.pt")
########################################################################################################

# Face detection and landmarks predictor setup
detector = dlib.get_frontal_face_detector()
predictor = dlib.shape_predictor("shape_predictor_68_face_landmarks.dat")

# Load the pretrained aging model
EXPERIMENT_TYPE = 'ffhq_aging'
EXPERIMENT_DATA_ARGS = {
    "ffhq_aging": {
        "model_path": sam_ffhq_aging,
        "transform": transforms.Compose([
            transforms.Resize((256, 256)),
            transforms.ToTensor(),
            transforms.Normalize([0.5, 0.5, 0.5], [0.5, 0.5, 0.5])
        ])
    }
}
EXPERIMENT_ARGS = EXPERIMENT_DATA_ARGS[EXPERIMENT_TYPE]
model_path = EXPERIMENT_ARGS['model_path']
ckpt = torch.load(model_path, map_location='cpu')
opts = ckpt['opts']
pprint.pprint(opts)
opts['checkpoint_path'] = model_path
opts = Namespace(**opts)
net = pSp(opts)
net.eval()
net.cuda()

print('Model successfully loaded!')

#######  Image quality checking func ######################

#def check_image_quality(image):
    ### Convert the image to grayscale
    ##gray_image = np.array(image.convert("L"))
    
    ### Check for under/over-exposure using histogram
    #hist = exposure.histogram(gray_image)
    #low_exposure = hist[0][:5].sum() > 0.5 * hist[0].sum()  # Significant pixels in dark range
    #high_exposure = hist[0][-5:].sum() > 0.5 * hist[0].sum()  # Significant pixels in bright range
    
    ### Check sharpness using Laplacian variance
    #sharpness = cv2.Laplacian(np.array(image), cv2.CV_64F).var()
    #low_sharpness = sharpness < 70  # Threshold for sharpness
    
    ### Check overall quality
    #if low_exposure or high_exposure or low_sharpness:
        #return False  # Image quality is insufficient
    #return True  # Image quality is sufficient

############################################################


# Functions for face and mouth region
def get_face_region(image):
    gray = cv2.cvtColor(np.array(image), cv2.COLOR_BGR2GRAY)
    faces = detector(gray)
    if len(faces) > 0:
        return faces[0]
    return None

def get_mouth_region(image):
    gray = cv2.cvtColor(np.array(image), cv2.COLOR_BGR2GRAY)
    faces = detector(gray)
    for face in faces:
        landmarks = predictor(gray, face)
        mouth_points = [(landmarks.part(i).x, landmarks.part(i).y) for i in range(48, 68)]
        return np.array(mouth_points, np.int32)
    return None


# Function to predict age

# old tensorflow function for age predict

def predict_age(image):
    image = np.array(image.resize((64, 64)))
    image = cv2.cvtColor(image, cv2.COLOR_RGB2GRAY) 
    image = image / 255.0  
    image = np.expand_dims(image, axis=0)

    ##### Predict age
    val = new_age_model.predict(np.array(image))
    age = val[0][0]
    return int(age)

#def predict_age(image):
    #age_calc_model.eval()
    ##### Load and preprocess the image
    #image = cv2.imread(image, cv2.IMREAD_GRAYSCALE)  # Load as grayscale
    #image = cv2.resize(image, (64, 64))  # Resize to 64x64
    #image = image / 255.0  # Normalize pixel values to [0, 1]
    #image = np.expand_dims(image, axis=0)  # Add batch dimension
    #image = np.expand_dims(image, axis=0)  # Add channel dimension
    #image = torch.tensor(image, dtype=torch.float32).to(device)

    # Convert to tensor
    #image_tensor = torch.tensor(image, dtype=torch.float32)

    #### Predict age
    #with torch.no_grad():
        #predicted_age = age_calc_model(image_tensor)
    
    #return int(predicted_age.item()) 

# Function for color correction
def color_correct(source, target):
    mean_src = np.mean(source, axis=(0, 1))
    std_src = np.std(source, axis=(0, 1))
    mean_tgt = np.mean(target, axis=(0, 1))
    std_tgt = np.std(target, axis=(0, 1))
    src_normalized = (source - mean_src) / std_src
    src_corrected = (src_normalized * std_tgt) + mean_tgt
    return np.clip(src_corrected, 0, 255).astype(np.uint8)

# Function to replace teeth
def replace_teeth(temp_image, aged_image):
    temp_image = np.array(temp_image)
    aged_image = np.array(aged_image)
    temp_mouth = get_mouth_region(temp_image)
    aged_mouth = get_mouth_region(aged_image)
    if temp_mouth is None or aged_mouth is None:
        return aged_image
    temp_mask = np.zeros_like(temp_image)
    cv2.fillConvexPoly(temp_mask, temp_mouth, (255, 255, 255))
    temp_mouth_region = cv2.bitwise_and(temp_image, temp_mask)
    temp_mouth_bbox = cv2.boundingRect(temp_mouth)
    aged_mouth_bbox = cv2.boundingRect(aged_mouth)
    temp_mouth_crop = temp_mouth_region[temp_mouth_bbox[1]:temp_mouth_bbox[1] + temp_mouth_bbox[3], temp_mouth_bbox[0]:temp_mouth_bbox[0] + temp_mouth_bbox[2]]
    temp_mask_crop = temp_mask[temp_mouth_bbox[1]:temp_mouth_bbox[1] + temp_mouth_bbox[3], temp_mouth_bbox[0]:temp_mouth_bbox[0] + temp_mouth_bbox[2]]
    temp_mouth_crop_resized = cv2.resize(temp_mouth_crop, (aged_mouth_bbox[2], aged_mouth_bbox[3]))
    temp_mask_crop_resized = cv2.resize(temp_mask_crop, (aged_mouth_bbox[2], aged_mouth_bbox[3]))
    aged_mouth_crop = aged_image[aged_mouth_bbox[1]:aged_mouth_bbox[1] + aged_mouth_bbox[3], aged_mouth_bbox[0]:aged_mouth_bbox[0] + aged_mouth_bbox[2]]
    temp_mouth_crop_resized = color_correct(temp_mouth_crop_resized, aged_mouth_crop)
    center = (aged_mouth_bbox[0] + aged_mouth_bbox[2] // 2, aged_mouth_bbox[1] + aged_mouth_bbox[3] // 2)
    seamless_teeth = cv2.seamlessClone(temp_mouth_crop_resized, aged_image, temp_mask_crop_resized, center, cv2.NORMAL_CLONE)
    return seamless_teeth

# Function to run alignment
def run_alignment(image):
    from scripts.align_all_parallel import align_face
    temp_image_path = "/tmp/temp_image.jpg"
    image.save(temp_image_path)
    aligned_image = align_face(filepath=temp_image_path, predictor=predictor)
    return aligned_image

# Function to apply aging
def apply_aging(image, target_age):
    img_transforms = EXPERIMENT_DATA_ARGS[EXPERIMENT_TYPE]['transform']
    input_image = img_transforms(image)
    age_transformers = [AgeTransformer(target_age=target_age)]
    results = []
    for age_transformer in age_transformers:
        with torch.no_grad():
            input_image_age = [age_transformer(input_image.cpu()).to('cuda')]
            input_image_age = torch.stack(input_image_age)
            result_tensor = net(input_image_age.float(), randomize_noise=False, resize=False)[0]
            result_image = tensor2im(result_tensor)
            results.append(np.array(result_image))
    final_result = results[0]
    return final_result

# Function to process the image
def process_image(uploaded_image):
    # Loading images for good and bad teeth
    temp_images_good = [Image.open(f"good_teeth/G{i}.JPG") for i in range(1, 4)]
    temp_images_bad = [Image.open(f"bad_teeth/B{i}.jpeg") for i in range(1, 5)]

    # Predicting the age
    predicted_age = predict_age(uploaded_image)

    if predicted_age >= 48:
        target_age =35+1

    else:
        target_age = predicted_age + 2

    # Aligning the face in the uploaded image
    aligned_image = run_alignment(uploaded_image)

    # Applying aging effect
    aged_image = apply_aging(aligned_image, target_age=target_age)

    # Randomly selecting teeth images
    good_teeth_image = temp_images_good[np.random.randint(0, len(temp_images_good))]
    bad_teeth_image = temp_images_bad[np.random.randint(0, len(temp_images_bad))]

    # Replacing teeth in aged image
    aged_image_good_teeth = replace_teeth(good_teeth_image, aged_image)
    aged_image_bad_teeth = replace_teeth(bad_teeth_image, aged_image)

    return aged_image_good_teeth, aged_image_bad_teeth, predicted_age, target_age

# Gradio Interface
def show_results(uploaded_image):
    ### Perform quality check
    #if not check_image_quality(uploaded_image):
        #return None, None, "Not_Allowed"

    # If quality is acceptable, continue with processing
    aged_image_good_teeth, aged_image_bad_teeth, predicted_age, target_age = process_image(uploaded_image)
    return aged_image_good_teeth, aged_image_bad_teeth, f"Predicted Age: {predicted_age}, Target Age: {target_age}"

iface = gr.Interface(
    fn=show_results,
    inputs=gr.Image(type="pil"),
    outputs=[gr.Image(type="pil"), gr.Image(type="pil"), gr.Textbox()],
    title="Aging Effect with Teeth Replacement",
    description="Upload an image to apply an aging effect. The application will generate two results: one with good teeth and one with bad teeth."
)

iface.launch()