api.py

from flask import Flask, request, jsonify
import numpy as np
import pickle
import os
import io
from tensorflow.keras.applications.mobilenet_v2 import MobileNetV2, preprocess_input
from tensorflow.keras.preprocessing import image
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import GlobalMaxPooling2D
from sklearn.neighbors import NearestNeighbors

app = Flask(__name__)

# Config
BASE_DIR = os.path.dirname(os.path.abspath(__file__))
INDEX_PATH = os.path.join(BASE_DIR, 'index', 'features.pkl')
FILES_PATH = os.path.join(BASE_DIR, 'index', 'filenames.pkl')
IMG_SIZE = (224, 224)

# Load Model & Index
base_model = MobileNetV2(weights='imagenet', include_top=False, input_shape=(224, 224, 3))
base_model.trainable = False
model = Sequential([base_model, GlobalMaxPooling2D()])

features = pickle.load(open(INDEX_PATH, 'rb'))
filenames = pickle.load(open(FILES_PATH, 'rb'))

neighbors = NearestNeighbors(n_neighbors=5, metric='euclidean', algorithm='brute')
neighbors.fit(features)

@app.route('/search', methods=['POST'])
def search():
    """
    Accepts an image upload and returns similar image filenames.
    """
    if 'image' not in request.files:
        return jsonify({"error": "No image uploaded"}), 400

    # Process Image
    file = request.files['image']
    img = image.load_img(io.BytesIO(file.read()), target_size=IMG_SIZE)
    img_array = image.img_to_array(img)
    expanded = np.expand_dims(img_array, axis=0)
    preprocessed = preprocess_input(expanded)

    # Extract Features
    query_vector = model.predict(preprocessed, verbose=0).flatten()
    query_vector = query_vector / np.linalg.norm(query_vector)

    # Find Neighbors
    distances, indices = neighbors.kneighbors([query_vector])

    results = []
    for i in range(len(indices[0])):
        idx = indices[0][i]
        results.append({
            "filename": filenames[idx],
            "score": float(1 - distances[0][i]) # Similarity Score
        })

    return jsonify(results)

if __name__ == '__main__':
    app.run(port=5001, debug=True)