Spaces:
Sleeping
Sleeping
File size: 9,094 Bytes
9a940be | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 | """
client/client_pipeline.py
SecureLens β TRUE Client-Side Pipeline
This is what runs on the PATIENT'S device (laptop, phone, hospital workstation).
It does:
1. Load and preprocess the X-ray
2. Extract 512-dim features using ResNet-18 backbone (LOCAL)
3. Encrypt the features using CKKS (LOCAL)
4. Send ONLY the ciphertext to the server
5. Receive encrypted result from server
6. Decrypt locally using secret key
THE SERVER NEVER SEES PLAINTEXT AT ANY POINT.
"""
import torch
import numpy as np
import tenseal as ts
import requests
import base64
import io
import os
import sys
from PIL import Image
from torchvision import transforms
sys.path.insert(0, os.path.join(os.path.dirname(__file__), ".."))
from crypto_layer.ckks_engine import CKKSEngine
class SecureLensClient:
"""
Full client-side pipeline for SecureLens.
Runs entirely on the patient's device.
"""
def __init__(
self,
model_path: str,
server_url: str = "http://127.0.0.1:7860",
):
self.server_url = server_url
# Step 1: Initialize CKKS engine (generates keys on client)
print("[Client] Initializing CKKS engine...")
self.ckks = CKKSEngine(
poly_modulus_degree=8192,
coeff_mod_bit_sizes=[60, 40, 40, 60],
global_scale=2**40,
)
# Step 2: Load ResNet-18 backbone (feature extractor)
print("[Client] Loading ResNet-18 backbone...")
self.backbone = self._load_backbone(model_path)
# Step 3: Image transform
self.transform = transforms.Compose([
transforms.Resize((224, 224)),
transforms.ToTensor(),
transforms.Normalize(
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]
),
])
print("[Client] Ready. Secret key on client only.")
def _load_backbone(self, model_path: str):
"""Loads ResNet-18 with classification head removed."""
from cloud_server.train_model import SecureLensNet
model = SecureLensNet(num_classes=2)
if os.path.exists(model_path):
model.load_state_dict(
torch.load(model_path, map_location="cpu"))
model.eval()
return model
# ββ Step A: Extract Features (CLIENT SIDE) βββββββββββββββββββββββββ
def extract_features(self, img_bytes: bytes) -> np.ndarray:
"""
Runs ResNet-18 backbone on the X-ray IMAGE.
Produces 512-dim feature vector.
RUNS ENTIRELY ON CLIENT DEVICE.
Raw image never leaves this function.
Args:
img_bytes: raw image bytes (PNG/JPG)
Returns:
numpy array shape (512,) β plaintext features
"""
img_pil = Image.open(io.BytesIO(img_bytes)).convert("RGB")
img_t = self.transform(img_pil).unsqueeze(0) # (1,3,224,224)
with torch.no_grad():
features = self.backbone.get_backbone_features(img_t)
feats = features.squeeze().numpy() # (512,)
print(f"[Client] Features extracted: shape={feats.shape}")
return feats
# ββ Step B: Encrypt Features (CLIENT SIDE) βββββββββββββββββββββββββ
def encrypt_features(self, features: np.ndarray) -> bytes:
"""
CKKS-encrypts the 512-dim feature vector.
RUNS ENTIRELY ON CLIENT DEVICE.
Returns ciphertext bytes ready for server.
Args:
features: numpy array (512,)
Returns:
bytes: serialized CKKS ciphertext (~326 KB)
"""
ct_bytes = self.ckks.encrypt_feature_vector_to_bytes(features)
print(f"[Client] Encrypted β {len(ct_bytes)//1024} KB ciphertext")
return ct_bytes
# ββ Step C: Send Ciphertext to Server ββββββββββββββββββββββββββββββ
def send_to_server(self, ct_bytes: bytes) -> bytes:
"""
Sends encrypted ciphertext to server.
Server receives ONLY this β no raw image, no plaintext features.
Args:
ct_bytes: serialized CKKS ciphertext
Returns:
bytes: server's encrypted result
"""
files = {"ciphertext": ("ct.bin", ct_bytes, "application/octet-stream")}
response = requests.post(
f"{self.server_url}/api/predict_encrypted",
files=files,
timeout=30,
)
response.raise_for_status()
data = response.json()
if data.get("error"):
raise ValueError(data["error"])
# Server returns encrypted result as base64
result_b64 = data["encrypted_result_b64"]
result_bytes = base64.b64decode(result_b64)
print(f"[Client] Server returned {len(result_bytes)//1024} KB encrypted result")
return result_bytes
# ββ Step D: Decrypt Result (CLIENT SIDE) βββββββββββββββββββββββββββ
def decrypt_result(self, enc_result_bytes: bytes) -> dict:
"""
Decrypts server's encrypted response.
RUNS ENTIRELY ON CLIENT DEVICE.
Uses secret key β which never left the client.
Args:
enc_result_bytes: encrypted logits from server
Returns:
dict: prediction, confidence, scores
"""
result = self.ckks.decrypt_prediction_from_bytes(enc_result_bytes)
print(f"[Client] Decrypted: {result['prediction']} "
f"({result['confidence']:.2%})")
return result
# ββ Full Pipeline βββββββββββββββββββββββββββββββββββββββββββββββββββ
def diagnose(self, img_bytes: bytes) -> dict:
"""
Full end-to-end SECURE pipeline:
raw image β features (client) β encrypt (client)
β ciphertext to server β HE inference (server)
β encrypted result to client β decrypt (client)
β diagnosis
SERVER NEVER SEES PLAINTEXT.
Args:
img_bytes: raw X-ray image bytes
Returns:
dict: prediction, confidence, scores, pipeline_steps
"""
import time
t0 = time.time()
print("\n[Client] Starting secure diagnosis pipeline...")
print("β" * 50)
# A: Extract features locally
print("[Client] Step 1: Extracting features locally...")
features = self.extract_features(img_bytes)
# B: Encrypt locally
print("[Client] Step 2: Encrypting with CKKS...")
ct_bytes = self.encrypt_features(features.copy())
# C: Send ciphertext only
print("[Client] Step 3: Sending ciphertext to server...")
enc_result_bytes = self.send_to_server(ct_bytes)
# D: Decrypt locally
print("[Client] Step 4: Decrypting result locally...")
result = self.decrypt_result(enc_result_bytes)
elapsed = (time.time() - t0) * 1000
print(f"[Client] Done in {elapsed:.0f}ms")
print("β" * 50)
result["latency_ms"] = round(elapsed, 1)
result["pipeline_steps"] = [
"X-ray loaded on client device",
"ResNet-18 extracted 512 features (client)",
"CKKS encrypted features β 326 KB ciphertext (client)",
"Ciphertext sent to cloud server",
"Server ran W1@enc(x)+b1, W2@enc(h)+b2 on ciphertext",
"Server returned encrypted logits",
"Client decrypted with secret key",
]
result["server_saw"] = "Ciphertext only β zero plaintext"
result["encryption_info"] = self.ckks.get_encryption_info()
return result
# ββ Command-line test βββββββββββββββββββββββββββββββββββββββββββββββββββ
if __name__ == "__main__":
import sys
if len(sys.argv) < 3:
print("Usage: python client_pipeline.py <model_path> <image_path>")
print("Example: python client_pipeline.py cloud_server/models/best_model.pth test.jpg")
sys.exit(1)
model_path = sys.argv[1]
image_path = sys.argv[2]
print("=" * 55)
print("SecureLens β Client Pipeline Test")
print("=" * 55)
client = SecureLensClient(
model_path=model_path,
server_url="http://127.0.0.1:7860",
)
with open(image_path, "rb") as f:
img_bytes = f.read()
result = client.diagnose(img_bytes)
print("\n" + "=" * 55)
print("RESULT")
print("=" * 55)
print(f" Prediction : {result['prediction']}")
print(f" Confidence : {result['confidence']:.2%}")
print(f" Normal : {result['normal_score']:.4f}")
print(f" Pneumonia : {result['pneumonia_score']:.4f}")
print(f" Latency : {result['latency_ms']}ms")
print(f" Server saw : {result['server_saw']}") |