app.py

import shutil
import subprocess
import sys
import os

def list_files_in_directory(directory):
    for root, dirs, files in os.walk(directory):
        for file in files:
            print(os.path.join(root, file))

list_files_in_directory("Data")

# Danh sách các gói cần cài đặt
required_packages = [
    "sentence-transformers",
    "faiss-cpu",  # Thay thế faiss-gpu bằng faiss-cpu để tránh lỗi - bạn có thể thay đổi tùy theo yêu cầu
    "numpy",  # Đảm bảo cài đặt numpy mới nhất
    "gradio",
    "matplotlib"
]

# Cài đặt các gói nếu chưa được cài đặt
def install_packages(packages):
    for package in packages:
        subprocess.check_call([sys.executable, "-m", "pip", "install", package])

install_packages(required_packages)

# Import các thư viện cần thiết sau khi cài đặt
from glob import glob
from PIL import Image
import numpy as np
import faiss
from sentence_transformers import SentenceTransformer
import matplotlib.pyplot as plt
import random
import gradio as gr

# Tạo embbeding
model = SentenceTransformer('clip-ViT-B-32')
image_path = "data"

# Tìm tất cả các tệp ảnh có định dạng .jpg
image_files = glob(os.path.join(image_path, "**/*.jpg"), recursive=True)
print("Found image files:", image_files)
if not image_files:
    raise ValueError("No image files found in the specified directory.")

chunk_size = 256

embeddings = []

def process_chunk(chunk):
    images = []
    for image_file in chunk:
        images.append(Image.open(image_file))
    chunk_embeddings = model.encode(images)
    return chunk_embeddings

for i in range(0, len(image_files), chunk_size):
    chunk = image_files[i:i + chunk_size]
    embeddings.extend(process_chunk(chunk))

print("Number of embeddings:", len(embeddings))

# Dựng lên vectorDB với FAISS
dimension = len(embeddings[0])
index = faiss.IndexFlatIP(dimension)
index = faiss.IndexIDMap(index)

vectors = np.array(embeddings).astype('float32')
index.add_with_ids(vectors, np.array(range(len(embeddings))))

# Save index vào file
faiss.write_index(index, "index.faiss")

# Ghi các tên ảnh vào tệp tin để load khi cần
with open("image_files.txt", "w") as f:
    for image_file in image_files:
        f.write(image_file + "\n")

def search_image(query, model, index, image_files, top_k=5):
    # Query có thể là ảnh hoặc text

    # Kiểm tra nếu query là một numpy array thì chuyển lại về dạng ảnh của PIL
    if isinstance(query, np.ndarray):
        query = Image.fromarray(query)
    query_embedding = model.encode(query)
    query_embedding = query_embedding.astype("float32").reshape(1, -1)

    distances, indices = index.search(query_embedding, top_k)

    retrieved_image_files = [image_files[i] for i in indices[0]]
    return retrieved_image_files

def display(query_image, query_text):
    if query_text:
        return search_image(query_text, model, index, image_files, 10)
    else:
        return search_image(query_image, model, index, image_files, 10)

with gr.Blocks() as demo:
    with gr.Row():
        with gr.Column():
            query_text = gr.Textbox(label="Nhập văn bản", lines=3)
            query_image = gr.Image(label="Tải lên ảnh", height=500)
        with gr.Column():
            result_images = gr.Gallery(label="Bộ ảnh kết quả", height=645, columns=2, rows=5)
    button = gr.Button("Tìm kiếm")

    button.click(fn=lambda query_image, query_text: display(query_image, query_text), inputs=[query_image, query_text], outputs=result_images)

demo.launch()