team14 / app.py
tagny's picture
team14: verio - working version 3 - clean gradio cache
c4700a2
raw
history blame contribute delete
No virus
25.8 kB
"""A local gradio app that detect matching images using FHE."""
import os
from pathlib import Path
import shutil
import time
from typing import Tuple
import requests
import numpy as np
import subprocess
import gradio as gr
from itertools import chain
import matplotlib.pyplot as plt
import matplotlib.image as img
import numpy as np
from PIL import Image
import torch
import torchvision.transforms as transforms
import torchvision.models as models
import cv2
from facenet_pytorch import InceptionResnetV1
from concrete.ml.deployment import FHEModelClient, FHEModelServer
from client_server_interface import FHEClient
from common import (
CLIENT_TMP_PATH,
ID_EXAMPLES,
SELFIE_EXAMPLES,
KEYS_PATH,
MATCHERS_PATH,
REPO_DIR,
SERVER_TMP_PATH,
SERVER_URL,
)
MODEL_PATH = "client_server"
# CLIENT_TMP_PATH = "client_tmp"
# Uncomment here to have both the server and client in the same terminal
subprocess.Popen(["uvicorn", "server:app"], cwd=REPO_DIR)
time.sleep(3)
# def decrypt_output_with_wrong_key(encrypted_image):
# """Decrypt the encrypted output using a different private key."""
# # Retrieve the matcher's deployment path
# matcher_path = MATCHERS_PATH / f"{matcher_name}/deployment"
# # Instantiate the client interface and generate a new private key
# wrong_client = FHEClient(matcher_path, matcher_name)
# wrong_client.generate_private_and_evaluation_keys(force=True)
# # Deserialize, decrypt and post-process the encrypted output using the new private key
# output_result = wrong_client.deserialize_decrypt_post_process(encrypted_image)
# # # For matchers that are expected to output black and white images, generate two other random
# # # channels for better display
# # if matcher_name in ["black and white", "ridge detection"]:
# # # Green channel
# # wrong_client.generate_private_and_evaluation_keys(force=True)
# # output_result[:, :, 1] = wrong_client.deserialize_decrypt_post_process(
# # encrypted_image
# # )[:, :, 0]
# # # Blue channel
# # wrong_client.generate_private_and_evaluation_keys(force=True)
# # output_result[:, :, 2] = wrong_client.deserialize_decrypt_post_process(
# # encrypted_image
# # )[:, :, 0]
# return output_result
def shorten_bytes_object(bytes_object, limit=500):
"""Shorten the input bytes object to a given length.
Encrypted data is too large for displaying it in the browser using Gradio. This function
provides a shorten representation of it.
Args:
bytes_object (bytes): The input to shorten
limit (int): The length to consider. Default to 500.
Returns:
str: Hexadecimal string shorten representation of the input byte object.
"""
# Define a shift for better display
shift = 100
return bytes_object[shift : limit + shift].hex()
def get_client():
"""Get the client API.
Args:
user_id (int): The current user's ID.
filter_name (str): The filter chosen by the user
Returns:
FHEClient: The client API.
"""
return FHEModelClient(MODEL_PATH)
def get_client_file_path(name, user_id):
"""Get the correct temporary file path for the client.
Args:
name (str): The desired file name.
user_id (int): The current user's ID.
filter_name (str): The filter chosen by the user
Returns:
pathlib.Path: The file path.
"""
return CLIENT_TMP_PATH / f"{name}_embedding_{user_id}"
def clean_temporary_files(n_keys=20):
"""Clean keys and encrypted images.
A maximum of n_keys keys and associated temporary files are allowed to be stored. Once this
limit is reached, the oldest files are deleted.
Args:
n_keys (int): The maximum number of keys and associated files to be stored. Default to 20.
"""
# Get the oldest key files in the key directory
key_dirs = sorted(KEYS_PATH.iterdir(), key=os.path.getmtime)
# If more than n_keys keys are found, remove the oldest
user_ids = []
if len(key_dirs) > n_keys:
n_keys_to_delete = len(key_dirs) - n_keys
for key_dir in key_dirs[:n_keys_to_delete]:
user_ids.append(key_dir.name)
shutil.rmtree(key_dir)
# Get all the encrypted objects in the temporary folder
client_files = Path(CLIENT_TMP_PATH).iterdir()
server_files = Path(SERVER_TMP_PATH).iterdir()
# Delete all files related to the ids whose keys were deleted
for file in chain(client_files, server_files):
for user_id in user_ids:
if user_id in file.name:
file.unlink()
def keygen(matcher_name):
"""Generate the private key associated to a matcher.
Args:
matcher_name (str): The current matcher to consider.
Returns:
(user_id, True) (Tuple[int, bool]): The current user's ID and a boolean used for visual display.
"""
# Clean temporary files
clean_temporary_files()
# Create an ID for the current user
user_id = np.random.randint(0, 2**32)
# user_id = 298147048
# Retrieve the client API
client = get_client()
# Generate a private key
client.generate_private_and_evaluation_keys(force=True)
# Retrieve the serialized evaluation key. In this case, as circuits are fully leveled, this
# evaluation key is empty. However, for software reasons, it is still needed for proper FHE
# execution
evaluation_key = client.get_serialized_evaluation_keys()
# Save evaluation_key as bytes in a file as it is too large to pass through regular Gradio
# buttons (see https://github.com/gradio-app/gradio/issues/1877)
evaluation_key_path = get_client_file_path("evaluation_key", user_id)
with evaluation_key_path.open("wb") as evaluation_key_file:
evaluation_key_file.write(evaluation_key)
return (user_id, True)
def detect_and_crop_face(
image: str,
min_aspect_ratio: float = 0.5,
max_aspect_ratio: float = 1.5,
min_face_size: float = 0.01,
max_face_size: float = 0.6,
) -> Tuple[np.ndarray, Tuple[int, int, int, int], np.ndarray]:
# Read the image
# image = cv2.imread(image_path)
image_path = "test"
if image is None:
print(f"Failed to load image: {image_path}")
return None
# Print the image depth to debug
print(f"Image Depth: {image.dtype}, Shape: {image.shape}")
# Check if the image is of type CV_64F (float64) and convert to uint8
if image.dtype == np.float64:
print(f"Converting image from float64 to uint8 for {image_path}")
image = cv2.convertScaleAbs(image) # Scale and convert to 8-bit
elif image.dtype != np.uint8:
print(f"Converting image from {image.dtype} to uint8 for {image_path}")
image = cv2.convertScaleAbs(image) # Convert to 8-bit unsigned
# Convert to grayscale
try:
gray_image = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
except cv2.error as e:
print(f"Error converting image to grayscale: {e} for {image_path}")
return None
# Load the face classifier
face_classifier = cv2.CascadeClassifier(
cv2.data.haarcascades + "haarcascade_frontalface_default.xml"
)
# Detect faces
faces = face_classifier.detectMultiScale(
gray_image,
scaleFactor=1.1,
minNeighbors=5,
minSize=(int(image.shape[1] * 0.1), int(image.shape[0] * 0.1)),
)
valid_faces = []
for x, y, w, h in faces:
aspect_ratio = w / h
face_area = w * h
image_area = image.shape[0] * image.shape[1]
face_size_ratio = face_area / image_area
if (
min_aspect_ratio <= aspect_ratio <= max_aspect_ratio
and min_face_size <= face_size_ratio <= max_face_size
):
valid_faces.append((x, y, w, h))
if not valid_faces:
print(f"No suitable faces detected in {image_path}")
return None
# Sort faces by area (descending) and select the largest
valid_faces.sort(key=lambda f: f[2] * f[3], reverse=True)
(x, y, w, h) = valid_faces[0]
# Crop the face
try:
face_crop = image[
int(y - h * 0.1) : int(y + h * 1.1), int(x - w * 0.1) : int(x + w * 1.1)
]
if face_crop.size == 0:
print(f"Failed to crop face for {image_path}: resulting crop is empty")
return None
except Exception as e:
print(f"Error cropping face from {image_path}: {e}")
return None
# Convert to RGB for display
try:
face_crop_rgb = cv2.cvtColor(face_crop, cv2.COLOR_BGR2RGB)
except cv2.error as e:
print(f"Error converting cropped face to RGB: {e} for {image_path}")
return None
return face_crop_rgb, (x, y, w, h), image
def preprocess_image(input_image):
# TODO change for facenet
model = InceptionResnetV1(pretrained="vggface2").eval()
input_image = np.array(input_image)
image_crop = detect_and_crop_face(image=input_image)
preprocess = transforms.Compose(
[
transforms.Resize((160, 160)), # Resize to 160x160 as required by the model
transforms.ToTensor(), # Convert to tensor
transforms.Normalize(
[0.5, 0.5, 0.5], [0.5, 0.5, 0.5]
), # Normalize to [-1, 1]
]
)
if image_crop[0] is not None:
img_tensor = preprocess(Image.fromarray(image_crop[0]))
img_tensor = img_tensor.unsqueeze(0)
with torch.no_grad():
embedding = model(img_tensor)
return embedding.numpy().flatten()
def encrypt(user_id, selfie_image, id_image):
"""Encrypt the given image for a specific user and filter.
Args:
user_id (int): The current user's ID.
selfie_image (np.ndarray): The image to encrypt.
id_image (np.ndarray): The image to encrypt.
Returns:
(input_image, encrypted_image_short) (Tuple[bytes]): The encrypted image and one of its
representation.
"""
if user_id == "":
raise gr.Error("Please generate the private key first.")
# for input_image in [selfie_image, id_image]:
# if input_image is None:
# raise gr.Error("Please choose an image first.")
# if input_image.shape[-1] != 3:
# raise ValueError(
# f"Input image must have 3 channels (RGB). Current shape: {input_image.shape}"
# )
# Resize the image if it hasn't the shape (100, 100, 3)
selfie_image_orig = selfie_image.copy()
id_image_orig = id_image.copy()
selfie_image = Image.fromarray(selfie_image).convert("RGB")
id_image = Image.fromarray(id_image).convert("RGB")
embeddings_selfie = preprocess_image(selfie_image)
embeddings_id = preprocess_image(id_image)
X = np.concatenate((embeddings_selfie, embeddings_id))[np.newaxis, ...]
# Retrieve the client API
client: FHEModelClient = get_client()
# Pre-process, encrypt and serialize the image
encrypted_image = client.quantize_encrypt_serialize(X)
# Save encrypted_image to bytes in a file, since too large to pass through regular Gradio
# buttons, https://github.com/gradio-app/gradio/issues/1877
encrypted_embedding = get_client_file_path("encrypted_embedding", user_id)
with encrypted_embedding.open("wb") as encrypted_image_file:
encrypted_image_file.write(encrypted_image)
# Create a truncated version of the encrypted image for display
encrypted_image_short = shorten_bytes_object(encrypted_image)
return (
encrypted_image_short,
resize_img(selfie_image_orig),
resize_img(id_image_orig),
)
def send_input(user_id):
"""Send the encrypted input image as well as the evaluation key to the server.
Args:
user_id (int): The current user's ID.
filter_name (str): The current filter to consider.
"""
# Get the evaluation key path
evaluation_key_path = get_client_file_path("evaluation_key", user_id)
if user_id == "" or not evaluation_key_path.is_file():
raise gr.Error("Please generate the private key first.")
encrypted_input_path = get_client_file_path("encrypted_embedding", user_id)
if not encrypted_input_path.is_file():
raise gr.Error(
"Please generate the private key and then encrypt an image first."
)
# Define the data and files to post
data = {
"user_id": user_id,
}
files = [
("files", open(encrypted_input_path, "rb")),
("files", open(evaluation_key_path, "rb")),
]
# Send the encrypted input image and evaluation key to the server
url = SERVER_URL + "send_input"
with requests.post(
url=url,
data=data,
files=files,
) as response:
return response.ok
def run_fhe(user_id):
"""Apply the filter on the encrypted image previously sent using FHE.
Args:
user_id (int): The current user's ID.
filter_name (str): The current filter to consider.
"""
data = {
"user_id": user_id,
}
# Trigger the FHE execution on the encrypted image previously sent
url = SERVER_URL + "run_fhe"
with requests.post(
url=url,
data=data,
) as response:
if response.ok:
return response.json()
else:
raise gr.Error("Please wait for the input image to be sent to the server.")
def get_output(user_id):
"""Retrieve the encrypted output image.
Args:
user_id (int): The current user's ID.
filter_name (str): The current filter to consider.
Returns:
encrypted_output_image_short (bytes): A representation of the encrypted result.
"""
data = {
"user_id": user_id,
}
# Retrieve the encrypted output image
url = SERVER_URL + "get_output"
with requests.post(
url=url,
data=data,
) as response:
if response.ok:
encrypted_output = response.content
# Save the encrypted output to bytes in a file as it is too large to pass through regular
# Gradio buttons (see https://github.com/gradio-app/gradio/issues/1877)
encrypted_output_path = get_client_file_path("encrypted_output", user_id)
with encrypted_output_path.open("wb") as encrypted_output_file:
encrypted_output_file.write(encrypted_output)
# # Decrypt the image using a different (wrong) key for display
# output_image_representation = decrypt_output_with_wrong_key(
# encrypted_output
# )
# return {
# encrypted_output_representation: gr.update(
# value=resize_img(output_image_representation)
# )
# }
# Create a truncated version of the encrypted image for display
encrypted_output_short = shorten_bytes_object(encrypted_output)
return encrypted_output_short
else:
raise gr.Error("Please wait for the FHE execution to be completed.")
def decrypt_output(user_id):
"""Decrypt the result.
Args:
user_id (int): The current user's ID.
filter_name (str): The current filter to consider.
Returns:
(output_image, False, False) ((Tuple[np.ndarray, bool, bool]): The decrypted output, as
well as two booleans used for resetting Gradio checkboxes
"""
if user_id == "":
raise gr.Error("Please generate the private key first.")
# Get the encrypted output path
encrypted_output_path = get_client_file_path("encrypted_output", user_id)
if not encrypted_output_path.is_file():
raise gr.Error("Please run the FHE execution first.")
# Load the encrypted output as bytes
with encrypted_output_path.open("rb") as encrypted_output_file:
encrypted_output = encrypted_output_file.read()
# Retrieve the client API
client = get_client()
# Deserialize, decrypt and post-process the encrypted output
decrypted_ouput = client.deserialize_decrypt_dequantize(encrypted_output)
print(f"Decrypted output: {decrypted_ouput.shape=}")
print(f"Decrypted output: {decrypted_ouput=}")
predicted_class_id = np.argmax(decrypted_ouput)
print(f"{predicted_class_id=}")
return "PASS" if predicted_class_id == 1 else "FAIL"
def resize_img(img, width=256, height=256):
"""Resize the image."""
if img.dtype != np.uint8:
img = img.astype(np.uint8)
img_pil = Image.fromarray(img)
# Resize the image
resized_img_pil = img_pil.resize((width, height))
# Convert back to a np array
return np.array(resized_img_pil)
# demo = gr.Blocks()
print("Starting the demo...")
with gr.Blocks(delete_cache=(60, 60)) as demo:
gr.Markdown(
"""
<!--p align="center">
<img width=200 src="https://user-images.githubusercontent.com/5758427/197816413-d9cddad3-ba38-4793-847d-120975e1da11.png">
</p-->
<h1 align="center">Verio “Privacy-Preserving Biometric Verification for Authentication”</h1>
<p align="center">
#ppaihackteam14
<a href="https://github.com/zama-ai/concrete-ml"> <img style="vertical-align: middle; display:inline-block; margin-right: 3px;" width=15 src="https://user-images.githubusercontent.com/5758427/197972109-faaaff3e-10e2-4ab6-80f5-7531f7cfb08f.png">Concrete-ML</a>
<a href="https://docs.zama.ai/concrete-ml"> <img style="vertical-align: middle; display:inline-block; margin-right: 3px;" width=15 src="https://user-images.githubusercontent.com/5758427/197976802-fddd34c5-f59a-48d0-9bff-7ad1b00cb1fb.png">Documentation</a>
<a href="https://zama.ai/community"> <img style="vertical-align: middle; display:inline-block; margin-right: 3px;" width=15 src="https://user-images.githubusercontent.com/5758427/197977153-8c9c01a7-451a-4993-8e10-5a6ed5343d02.png">Community</a>
<a href="https://twitter.com/zama_fhe"> <img style="vertical-align: middle; display:inline-block; margin-right: 3px;" width=15 src="https://user-images.githubusercontent.com/5758427/197975044-bab9d199-e120-433b-b3be-abd73b211a54.png">@zama_fhe</a>
</p>
<!--p align="center">
<img src="https://user-images.githubusercontent.com/56846628/219605302-5baafac4-cf6f-4f06-9a96-91cef2b84a63.png" width="70%" height="70%">
</p-->
"""
)
gr.Markdown("## Client side")
gr.Markdown("### Step 1: Upload input images. ")
# gr.Markdown(
# f"The image will automatically be resized to shape ({INPUT_SHAPE[0]}x{INPUT_SHAPE[1]}). "
# "The image here, however, is displayed in its original resolution. The true image used "
# "in this demo can be seen in Step 8."
# )
gr.Markdown("The query image to certify.")
with gr.Row():
input_query_img = gr.Image(
value=None,
label="Upload an image here.",
height=256,
width=256,
# source="upload",
interactive=True,
)
selfie_examples = gr.Examples(
examples=SELFIE_EXAMPLES,
inputs=[input_query_img],
examples_per_page=5,
label="Examples to use.",
)
gr.Markdown("The reference image.")
with gr.Row():
input_reference_img = gr.Image(
value=None,
label="Upload an image here.",
height=256,
width=256,
# source="upload",
interactive=True,
)
id_examples = gr.Examples(
examples=ID_EXAMPLES,
inputs=[input_reference_img],
examples_per_page=5,
label="Examples to use.",
)
# gr.Markdown("### Step 2: Choose your matcher.")
# matcher_name = gr.Dropdown(
# choices=AVAILABLE_MATCHERS,
# value="random guessing",
# label="Choose your matcher",
# interactive=True,
# )
# gr.Markdown("#### Notes")
# gr.Markdown(
# """
# - The private key is used to encrypt and decrypt the data and will never be shared.
# - No public key is required for these matcher operators.
# """
# )
gr.Markdown("### Step 3: Generate the private key.")
keygen_button = gr.Button("Generate the private key.")
with gr.Row():
keygen_checkbox = gr.Checkbox(label="Private key generated:", interactive=False)
user_id = gr.Textbox(label="", max_lines=2, interactive=False, visible=False)
# encrypted_query_image = gr.Textbox(
# value=ENCRYPTED_QUERY_NAME,
# label="",
# max_lines=2,
# interactive=False,
# visible=False,
# )
# encrypted_reference_image = gr.Textbox(
# value=ENCRYPTED_REFERENCE_NAME,
# label="",
# max_lines=2,
# interactive=False,
# visible=False,
# )
gr.Markdown("### Step 4: Encrypt the input images using FHE.")
encrypt_button = gr.Button("Encrypt the images using FHE.")
with gr.Row():
encrypted_input = gr.Textbox(
label="Encrypted input images representation:",
max_lines=2,
interactive=False,
)
gr.Markdown("## Server side")
gr.Markdown(
"The encrypted value is received by the server. The server can then compute the matcher "
"directly over encrypted values. Once the computation is finished, the server returns "
"the encrypted results to the client."
)
gr.Markdown("### Step 5: Send the encrypted images to the server.")
send_input_button = gr.Button("Send the encrypted images to the server.")
send_input_checkbox = gr.Checkbox(label="Encrypted images sent.", interactive=False)
gr.Markdown("### Step 6: Run FHE execution.")
execute_fhe_button = gr.Button("Run FHE execution.")
fhe_execution_time = gr.Textbox(
label="Total FHE execution time (in seconds):", max_lines=1, interactive=False
)
gr.Markdown("### Step 7: Receive the encrypted output from the server.")
gr.Markdown(
"The result displayed here is the encrypted result sent by the server, which has been "
"decrypted using a different private key. This is only used to visually represent an "
"encrypted result."
)
get_output_button = gr.Button(
"Receive the encrypted output result from the server."
)
with gr.Row():
# encrypted_output_representation = gr.Label()
encrypted_output_representation = gr.Textbox(
label="Encrypted encrypted output result:",
max_lines=2,
interactive=False,
)
# encrypted_output_representation = gr.Image(
# label=f"Encrypted output representation ({INPUT_SHAPE[0]}x{INPUT_SHAPE[1]}):",
# interactive=False,
# height=256,
# width=256,
# )
gr.Markdown("## Client side")
gr.Markdown(
"The encrypted output is sent back to the client, who can finally decrypt it with the "
"private key. Only the client is aware of the original input images and the result of the matching."
)
gr.Markdown("### Step 8: Decrypt the output.")
gr.Markdown(
"The images displayed on the left are the input images used during the demo. The output result "
"can be seen on the right."
)
decrypt_button = gr.Button("Decrypt the output")
# Final input vs output display
with gr.Row():
original_query_image = gr.Image(
input_query_img.value,
label=f"Input query image:",
interactive=False,
height=256,
width=256,
)
original_reference_image = gr.Image(
input_reference_img.value,
label=f"Input reference image:",
interactive=False,
height=256,
width=256,
)
output_result = gr.Label()
# output_image = gr.Image(
# label=f"Output image ({INPUT_SHAPE[0]}x{INPUT_SHAPE[1]}):",
# interactive=False,
# height=256,
# width=256,
# )
# Button to generate the private key
keygen_button.click(
keygen,
inputs=[],
outputs=[user_id, keygen_checkbox],
)
# Button to encrypt input query on the client side
encrypt_button.click(
encrypt,
inputs=[user_id, input_query_img, input_reference_img],
outputs=[encrypted_input, original_query_image, original_reference_image],
)
# Button to send the encodings to the server using post method
send_input_button.click(send_input, inputs=[user_id], outputs=[send_input_checkbox])
# Button to send the encodings to the server using post method
execute_fhe_button.click(run_fhe, inputs=[user_id], outputs=[fhe_execution_time])
# Button to send the encodings to the server using post method
get_output_button.click(
get_output,
inputs=[user_id],
outputs=[encrypted_output_representation],
)
# Button to decrypt the output on the client side
decrypt_button.click(
decrypt_output,
inputs=[user_id],
# outputs=[output_result, original_query_image, original_reference_image],
outputs=[output_result],
)
gr.Markdown(
"The app was built with [Concrete-ML](https://github.com/zama-ai/concrete-ml), a "
"Privacy-Preserving Machine Learning (PPML) open-source set of tools by [Zama](https://zama.ai/). "
"Try it yourself and don't forget to star on Github &#11088;."
)
demo.launch(share=False)