app.py

# -*- coding:UTF-8 -*-
#!/usr/bin/env python
import numpy as np
import gradio as gr
import roop.globals
from roop.core import (
    start,
    decode_execution_providers,
    suggest_max_memory,
    suggest_execution_threads,
)
from roop.processors.frame.core import get_frame_processors_modules
from roop.utilities import normalize_output_path
import os
from PIL import Image
from datetime import datetime
from huggingface_hub import HfApi, login
from datasets import load_dataset, Dataset
import json
import shutil
from dotenv import load_dotenv
import cv2
from insightface.app import FaceAnalysis

# Load environment variables
load_dotenv()

# Hàm tính cosine similarity để mày so sánh "điểm tương đồng" của khuôn mặt
def cosine_similarity(a, b):
    return np.dot(a, b) / (np.linalg.norm(a) * np.linalg.norm(b) + 1e-6)

# Class FaceIntegrDataset nguyên bản (cho image swap, không cần "xịn" cho video)
class FaceIntegrDataset:
    def __init__(self, repo_id="Arrcttacsrks/face_integrData"):
        self.token = os.getenv('hf_token')
        if not self.token:
            raise ValueError("HF_TOKEN environment variable is not set")
        self.repo_id = repo_id
        self.api = HfApi()
        login(self.token)
        self.temp_dir = "temp_dataset"
        os.makedirs(self.temp_dir, exist_ok=True)

    def create_date_folder(self):
        current_date = datetime.now().strftime("%Y-%m-%d")
        folder_path = os.path.join(self.temp_dir, current_date)
        os.makedirs(folder_path, exist_ok=True)
        return folder_path, current_date

    def save_metadata(self, source_path, target_path, output_path, timestamp):
        metadata = {
            "timestamp": timestamp,
            "source_image": source_path,
            "target_image": target_path,
            "output_image": output_path,
            "date_created": datetime.now().strftime("%Y-%m-%d %H:%M:%S")
        }
        return metadata

    def upload_to_hf(self, local_folder, date_folder):
        try:
            self.api.upload_folder(
                folder_path=local_folder,
                repo_id=self.repo_id,
                repo_type="dataset",
                path_in_repo=date_folder
            )
            return True
        except Exception as e:
            print(f"Error uploading to Hugging Face: {str(e)}")
            return False

# Hàm swap_face nguyên bản dành cho ghép ảnh tĩnh
def swap_face(source_file, target_file, doFaceEnhancer):
    folder_path = None
    try:
        dataset_handler = FaceIntegrDataset()
        folder_path, date_folder = dataset_handler.create_date_folder()
        timestamp = datetime.now().strftime("%S-%M-%H-%d-%m-%Y")
        source_path = os.path.join(folder_path, f"source_{timestamp}.jpg")
        target_path = os.path.join(folder_path, f"target_{timestamp}.jpg")
        output_path = os.path.join(folder_path, f"OutputImage{timestamp}.jpg")

        if source_file is None or target_file is None:
            raise ValueError("Source and target images are required")
            
        Image.fromarray(source_file).save(source_path)
        Image.fromarray(target_file).save(target_path)
        
        print("source_path: ", source_path)
        print("target_path: ", target_path)
        
        roop.globals.source_path = source_path
        roop.globals.target_path = target_path
        roop.globals.output_path = normalize_output_path(
            roop.globals.source_path, 
            roop.globals.target_path, 
            output_path
        )
        
        if doFaceEnhancer:
            roop.globals.frame_processors = ["face_swapper", "face_enhancer"]
        else:
            roop.globals.frame_processors = ["face_swapper"]
        
        roop.globals.headless = True
        roop.globals.keep_fps = True
        roop.globals.keep_audio = True
        roop.globals.keep_frames = False
        roop.globals.many_faces = False
        roop.globals.video_encoder = "libx264"
        roop.globals.video_quality = 18
        roop.globals.max_memory = suggest_max_memory()
        roop.globals.execution_providers = decode_execution_providers(["cuda"])
        roop.globals.execution_threads = suggest_execution_threads()
        
        print(
            "start process",
            roop.globals.source_path,
            roop.globals.target_path,
            roop.globals.output_path,
        )
        
        for frame_processor in get_frame_processors_modules(roop.globals.frame_processors):
            if not frame_processor.pre_check():
                return None
        
        start()
        
        metadata = dataset_handler.save_metadata(
            f"source_{timestamp}.jpg",
            f"target_{timestamp}.jpg",
            f"OutputImage{timestamp}.jpg",
            timestamp
        )
        
        metadata_path = os.path.join(folder_path, f"metadata_{timestamp}.json")
        with open(metadata_path, 'w') as f:
            json.dump(metadata, f, indent=4)
        
        upload_success = dataset_handler.upload_to_hf(folder_path, date_folder)
        
        if upload_success:
            print(f"Successfully uploaded files to dataset {dataset_handler.repo_id}")
        else:
            print("Failed to upload files to Hugging Face dataset")
        
        if os.path.exists(output_path):
            output_image = Image.open(output_path)
            output_array = np.array(output_image)
            shutil.rmtree(folder_path)
            return output_array
        else:
            print("Output image not found")
            if folder_path and os.path.exists(folder_path):
                shutil.rmtree(folder_path)
            return None
        
    except Exception as e:
        print(f"Error in face swap process: {str(e)}")
        if folder_path and os.path.exists(folder_path):
            shutil.rmtree(folder_path)
        raise gr.Error(f"Face swap failed: {str(e)}")

# HÀM MỚI: Xử lý ghép mặt cho 1 frame video bằng cách "mượn" thuật toán của roop
def swap_face_frame(frame_bgr, replacement_face_rgb, doFaceEnhancer):
    # Tao convert frame từ BGR sang RGB vì PIL làm việc với RGB – không cho mày chê!
    frame_rgb = cv2.cvtColor(frame_bgr, cv2.COLOR_BGR2RGB)
    temp_dir = "temp_faceswap_frame"
    os.makedirs(temp_dir, exist_ok=True)
    timestamp = datetime.now().strftime("%S-%M-%H-%d-%m-%Y")
    source_path = os.path.join(temp_dir, f"source_{timestamp}.jpg")
    target_path = os.path.join(temp_dir, f"target_{timestamp}.jpg")
    output_path = os.path.join(temp_dir, f"OutputImage_{timestamp}.jpg")
    Image.fromarray(frame_rgb).save(source_path)
    Image.fromarray(replacement_face_rgb).save(target_path)
    
    roop.globals.source_path = source_path
    roop.globals.target_path = target_path
    roop.globals.output_path = normalize_output_path(source_path, target_path, output_path)
    
    if doFaceEnhancer:
        roop.globals.frame_processors = ["face_swapper", "face_enhancer"]
    else:
        roop.globals.frame_processors = ["face_swapper"]
        
    roop.globals.headless = True
    roop.globals.keep_fps = True
    roop.globals.keep_audio = True
    roop.globals.keep_frames = False
    roop.globals.many_faces = False
    roop.globals.video_encoder = "libx264"
    roop.globals.video_quality = 18
    roop.globals.max_memory = suggest_max_memory()
    roop.globals.execution_providers = decode_execution_providers(["cuda"])
    roop.globals.execution_threads = suggest_execution_threads()
    
    start()
    
    if os.path.exists(output_path):
        swapped_img = np.array(Image.open(output_path))
    else:
        swapped_img = frame_rgb
    shutil.rmtree(temp_dir)
    return swapped_img

# HÀM MỚI: Xử lý ghép mặt cho video frame-by-frame với insightface để so sánh khuôn mặt
def swap_face_video(reference_face, replacement_face, video_input, similarity_threshold, doFaceEnhancer):
    """
    reference_face: Ảnh tham chiếu (RGB) để khóa khuôn mặt
    replacement_face: Ảnh ghép (RGB)
    video_input: Đường dẫn file video đầu vào
    similarity_threshold: Ngưỡng (0.0 - 1.0) cho tỉ lệ tương đồng
    doFaceEnhancer: Boolean, có áp dụng cải thiện chất lượng hay không
    """
    try:
        # Chuẩn bị insightface
        fa = FaceAnalysis()
        fa.prepare(ctx_id=0, nms=0.4)
        
        # Lấy embedding của khuôn mặt tham chiếu
        ref_detections = fa.get(reference_face)
        if not ref_detections:
            raise gr.Error("Không phát hiện khuôn mặt trong ảnh tham chiếu!")
        ref_embedding = ref_detections[0].embedding
        
        # Mở video đầu vào
        cap = cv2.VideoCapture(video_input)
        if not cap.isOpened():
            raise gr.Error("Không mở được video đầu vào!")
        fps = cap.get(cv2.CAP_PROP_FPS)
        width = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))
        height = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
        
        output_video_path = "temp_faceswap_video.mp4"
        fourcc = cv2.VideoWriter_fourcc(*'mp4v')
        out = cv2.VideoWriter(output_video_path, fourcc, fps, (width, height))
        
        frame_index = 0
        while True:
            ret, frame = cap.read()
            if not ret:
                break
            # Convert frame sang RGB để insightface xử lý
            frame_rgb = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
            detections = fa.get(frame_rgb)
            swap_this_frame = False
            for det in detections:
                sim = cosine_similarity(det.embedding, ref_embedding)
                if sim >= similarity_threshold:
                    swap_this_frame = True
                    break
            if swap_this_frame:
                # Ghép mặt từ replacement_face vào frame
                swapped_frame_rgb = swap_face_frame(frame, replacement_face, doFaceEnhancer)
                # Convert ngược lại sang BGR để ghi video
                swapped_frame = cv2.cvtColor(swapped_frame_rgb, cv2.COLOR_RGB2BGR)
            else:
                swapped_frame = frame
            out.write(swapped_frame)
            frame_index += 1
            print(f"Đã xử lý frame {frame_index}")
        cap.release()
        out.release()
        return output_video_path
    except Exception as e:
        print(f"Lỗi khi xử lý video: {str(e)}")
        raise gr.Error(f"Face swap video failed: {str(e)}")

# Giao diện Gradio được "đấm" thêm mục chuyển đổi giữa image và video
def create_interface():
    custom_css = """
    .container {
        max-width: 1200px;
        margin: auto;
        padding: 20px;
    }
    .output-image {
        min-height: 400px;
        border: 1px solid #ccc;
        border-radius: 8px;
        padding: 10px;
    }
    """
    title = "Face - Integrator"
    description = r"""
    Upload source and target images to perform face swap.
    """
    article = r"""
    <div style="text-align: center; max-width: 650px; margin: 40px auto;">
        <p>
            This tool performs face swapping with optional enhancement.
        </p>
    </div>
    """
    with gr.Blocks(title=title, css=custom_css) as app:
        gr.Markdown(f"<h1 style='text-align: center;'>{title}</h1>")
        gr.Markdown(description)
        with gr.Tabs():
            with gr.TabItem("FaceSwap Image"):
                with gr.Row():
                    with gr.Column(scale=1):
                        source_image = gr.Image(
                            label="Source Image",
                            type="numpy",
                            sources=["upload"]
                        )
                    with gr.Column(scale=1):
                        target_image = gr.Image(
                            label="Target Image",
                            type="numpy",
                            sources=["upload"]
                        )
                    with gr.Column(scale=1):
                        output_image = gr.Image(
                            label="Output Image",
                            type="numpy",
                            interactive=False,
                            elem_classes="output-image"
                        )
                with gr.Row():
                    enhance_checkbox = gr.Checkbox(
                        label="Apply the algorithm?",
                        info="Image Quality Improvement",
                        value=False
                    )
                with gr.Row():
                    process_btn = gr.Button(
                        "Process Face Swap",
                        variant="primary",
                        size="lg"
                    )
                process_btn.click(
                    fn=swap_face,
                    inputs=[source_image, target_image, enhance_checkbox],
                    outputs=output_image,
                    api_name="swap_face"
                )
            with gr.TabItem("FaceSwap Video"):
                gr.Markdown("<h2 style='text-align:center;'>FaceSwap Video</h2>")
                with gr.Row():
                    ref_image = gr.Image(
                        label="Ảnh mặt tham chiếu (khóa khuôn mặt)",
                        type="numpy",
                        sources=["upload"]
                    )
                    swap_image = gr.Image(
                        label="Ảnh mặt ghép",
                        type="numpy",
                        sources=["upload"]
                    )
                video_input = gr.Video(
                    label="Video đầu vào",
                    type="filepath"
                )
                similarity_threshold = gr.Slider(
                    minimum=0.0,
                    maximum=1.0,
                    step=0.01,
                    value=0.7,
                    label="Tỉ lệ tương đồng"
                )
                enhance_checkbox_video = gr.Checkbox(
                    label="Áp dụng cải thiện chất lượng ảnh",
                    info="Tùy chọn cải thiện",
                    value=False
                )
                process_video_btn = gr.Button(
                    "Xử lý FaceSwap Video",
                    variant="primary",
                    size="lg"
                )
                video_output = gr.Video(
                    label="Video kết quả",
                    type="filepath"
                )
                process_video_btn.click(
                    fn=swap_face_video,
                    inputs=[ref_image, swap_image, video_input, similarity_threshold, enhance_checkbox_video],
                    outputs=video_output,
                    api_name="swap_face_video"
                )
        gr.Markdown(article)
    return app

def main():
    app = create_interface()
    app.launch(share=False)

if __name__ == "__main__":
    main()