chore: clean and fix typos

README.md

@@ -41,15 +41,27 @@ pip3 install -U pip wheel setuptools --ignore-installed
 pip3 install -r requirements.txt --ignore-installed
 ```
 
-If not on Linux, or if you want to compile the FHE filters by yourself:
+If you are not on Linux, you will need to generate the files manually before launching the app. 
+See section [Generating new files](#generating-new-files).
+
+
+## Run the following steps each time you relaunch the application
+
+In a terminal, run:
 
 <!--pytest-codeblocks:skip-->
 
 ```bash
-python3 compile.py
+source .venv/bin/activate
+python3 app.py
 ```
 
-Check it finish well (with a "Done!").
+## Interacting with the application
+
+Open the given URL link (search for a line like `Running on local URL:  http://127.0.0.1:8888/`).
+
+
+## Generating new files
 
 It is also possible to manually add some new filters in `filters.py`. Yet, in order to be able to use
 them interactively in the app, you first need to update the `AVAILABLE_FILTERS` list found in `common.py`
@@ -61,17 +73,5 @@ and then compile them by running :
 python3 generate_dev_filters.py
 ```
 
-## Run the following steps each time you relaunch the application
+Check it finishes well (by printing "Done!").
 
-In a terminal, run:
-
-<!--pytest-codeblocks:skip-->
-
-```bash
-source .venv/bin/activate
-python3 app.py
-```
-
-## Interacting with the application
-
-Open the given URL link (search for a line like `Running on local URL:  http://127.0.0.1:8888/`).

app.py

@@ -33,7 +33,7 @@ def decrypt_output_with_wrong_key(encrypted_image, filter_name):
     filter_path = FILTERS_PATH / f"{filter_name}/deployment"
 
     # Instantiate the client interface and generate a new private key
-    wrong_client = FHEClient(filter_path, WRONG_KEYS_PATH)
+    wrong_client = FHEClient(filter_path, WRONG_KEYS_PATH, filter_name)
     wrong_client.generate_private_and_evaluation_keys(force=True)
 
     # Deserialize, decrypt and post-process the encrypted output using the new private key
@@ -72,7 +72,9 @@ def get_client(user_id, filter_name):
         FHEClient: The client API.
     """
     return FHEClient(
-        FILTERS_PATH / f"{filter_name}/deployment", KEYS_PATH / f"{filter_name}_{user_id}"
+        FILTERS_PATH / f"{filter_name}/deployment", 
+        KEYS_PATH / f"{filter_name}_{user_id}", 
+        filter_name,
     )
 
 

client_server_interface.py

@@ -1,10 +1,9 @@
 "Client-server interface custom implementation for filter models."
 
-import zipfile
-import json
+import concrete.numpy as cnp
+
 from filters import Filter
 
-import concrete.numpy as cnp
 
 class FHEServer:
     """Server interface run a FHE circuit."""
@@ -31,20 +30,22 @@ class FHEServer:
             bytes: The filter's output.
         """
         # Deserialize the encrypted input image and the evaluation keys
-        deserialized_encrypted_image = self.server.client_specs.unserialize_public_args(
+        encrypted_image = self.server.client_specs.unserialize_public_args(
             serialized_encrypted_image
         )
-        deserialized_evaluation_keys = cnp.EvaluationKeys.unserialize(serialized_evaluation_keys)
+        evaluation_keys = cnp.EvaluationKeys.unserialize(serialized_evaluation_keys)
 
         # Execute the filter in FHE
-        result = self.server.run(
-            deserialized_encrypted_image, deserialized_evaluation_keys
+        encrypted_output = self.server.run(
+            encrypted_image, evaluation_keys
         )
 
         # Serialize the encrypted output image
-        serialized_result = self.server.client_specs.serialize_public_result(result)
+        serialized_encrypted_output = self.server.client_specs.serialize_public_result(
+            encrypted_output
+        )
 
-        return serialized_result
+        return serialized_encrypted_output
 
 
 class FHEDev:
@@ -70,13 +71,6 @@ class FHEDev:
             "The model must be compiled before saving it."
         )
 
-        # Export to json the parameters needed for loading the filter in the other interfaces
-        serialized_processing = {"filter_name": self.filter.filter_name}
-
-        json_path = self.path_dir / "serialized_processing.json"
-        with open(json_path, "w", encoding="utf-8") as file:
-            json.dump(serialized_processing, file)
-
         # Save the circuit for the server
         path_circuit_server = self.path_dir / "server.zip"
         self.filter.fhe_circuit.server.save(path_circuit_server)
@@ -85,19 +79,16 @@ class FHEDev:
         path_circuit_client = self.path_dir / "client.zip"
         self.filter.fhe_circuit.client.save(path_circuit_client)
 
-        with zipfile.ZipFile(path_circuit_client, "a") as zip_file:
-            zip_file.write(filename=json_path, arcname="serialized_processing.json")
-
 
 class FHEClient:
     """Client interface to encrypt and decrypt FHE data associated to a Filter."""
 
-    def __init__(self, path_dir, key_dir):
+    def __init__(self, path_dir, key_dir, filter_name):
         """Initialize the FHE interface.
 
         Args:
-            path_dir (Path): the path to the directory where the circuit is saved
-            key_dir (Path): the path to the directory where the keys are stored
+            path_dir (Path): The path to the directory where the circuit is saved.
+            key_dir (Path): The path to the directory where the keys are stored.
         """
         self.path_dir = path_dir
         self.key_dir = key_dir
@@ -108,13 +99,7 @@ class FHEClient:
         # Load the client
         self.client = cnp.Client.load(self.path_dir / "client.zip", self.key_dir)
 
-        # Load the parameters
-        with zipfile.ZipFile(self.path_dir / "client.zip") as client_zip:
-            with client_zip.open("serialized_processing.json", mode="r") as file:
-                serialized_processing = json.load(file)
-
         # Instantiate the filter
-        filter_name = serialized_processing["filter_name"]
         self.filter = Filter(filter_name)
 
     def generate_private_and_evaluation_keys(self, force=False):
@@ -134,7 +119,7 @@ class FHEClient:
         return self.client.evaluation_keys.serialize()
 
     def pre_process_encrypt_serialize(self, input_image):
-        """Pre-process, encrypt and serialize the input image.
+        """Pre-process, encrypt and serialize the input image in the clear.
 
         Args:
             input_image (numpy.ndarray): The image to pre-process, encrypt and serialize.
@@ -153,7 +138,7 @@ class FHEClient:
         return serialized_encrypted_image
 
     def deserialize_decrypt_post_process(self, serialized_encrypted_output_image):
-        """Deserialize, decrypt and post-process the output image.
+        """Deserialize, decrypt and post-process the output image in the clear.
 
         Args:
             serialized_encrypted_output_image (bytes): The serialized and encrypted output image.

compile.py

@@ -1,45 +0,0 @@
-"A script to manually compile all filters."
-
-import json
-import shutil
-import onnx
-
-from common import AVAILABLE_FILTERS, FILTERS_PATH, KEYS_PATH
-from client_server_interface import FHEClient, FHEDev
-
-print("Starting compiling the filters.")
-
-for filter_name in AVAILABLE_FILTERS:
-    print("\nCompiling filter:", filter_name)
-
-    # Retrieve the deployment files associated to the current filter
-    deployment_path = FILTERS_PATH / f"{filter_name}/deployment"
-
-    # Retrieve the client associated to the current filter
-    filter = FHEClient(deployment_path, KEYS_PATH).filter
-
-    # Load the onnx graph
-    onnx_graph = onnx.load(FILTERS_PATH / f"{filter_name}/server.onnx")
-
-    # Compile the filter on a representative inputset, using the loaded onnx graph
-    filter.compile(onnx_graph=onnx_graph)
-
-    processing_json_path = deployment_path / "serialized_processing.json"
-
-    # Load the serialized_processing.json file
-    with open(processing_json_path, "r") as f:
-        serialized_processing = json.load(f)
-
-    # Delete the deployment folder and its content if it exist
-    if deployment_path.is_dir():
-        shutil.rmtree(deployment_path)
-
-    # Save the development files needed for deployment
-    fhe_dev = FHEDev(filter, deployment_path)
-    fhe_dev.save()
-
-    # Write the serialized_processing.json file in the deployment directory
-    with open(processing_json_path, "w") as f:
-        json.dump(serialized_processing, f)
-
-print("Done!")

filters.py

@@ -7,7 +7,6 @@ from common import AVAILABLE_FILTERS, INPUT_SHAPE
 
 from concrete.numpy.compilation.compiler import Compiler
 from concrete.ml.common.utils import generate_proxy_function
-from concrete.ml.onnx.convert import get_equivalent_numpy_forward
 from concrete.ml.torch.numpy_module import NumpyModule
 
 
@@ -57,7 +56,7 @@ class TorchRotate(nn.Module):
 
 
 class TorchConv(nn.Module):
-    """Torch model for applying convolution operators on images."""
+    """Torch model with a single convolution operator."""
 
     def __init__(self, kernel, n_in_channels=3, n_out_channels=3, groups=1, threshold=None):
         """Initialize the filter.
@@ -73,14 +72,13 @@ class TorchConv(nn.Module):
         self.threshold = threshold
 
     def forward(self, x):
-        """Forward pass for filtering the image using a 1D or 2D kernel.
+        """Forward pass with a single convolution using a 1D or 2D kernel.
 
         Args:
             x (torch.Tensor): The input image.
 
         Returns:
             torch.Tensor: The filtered image.
-
         """
         # Define the convolution parameters
         stride = 1
@@ -207,43 +205,36 @@ class Filter:
             # ususally displayed as such
             self.repeat_out_channels = True
 
-    def compile(self, onnx_graph=None):
-        """Compile the model on a representative inputset.
-
-        Args:
-            onnx_graph (onnx.ModelProto): The loaded onnx model to consider. If None, it will be
-                generated automatically using a NumpyModule. Default to None.
-        """
+    def compile(self):
+        """Compile the filter on a representative inputset."""
         # Generate a random representative set of images used for compilation, following Torch's 
         # shape format (batch, in_channels, image_height, image_width) for each samples
         # This version's compiler only handles tuples of 1-batch array as inputset, meaning we need
         # to define the inputset as a Tuple[np.ndarray[shape=(1, 3, H, W)]]  
         np.random.seed(42)
         inputset = tuple(
-            np.random.randint(0, 255, size=((1, 3) + INPUT_SHAPE), dtype=np.int64) for _ in range(10)
+            np.random.randint(0, 256, size=((1, 3) + INPUT_SHAPE), dtype=np.int64) for _ in range(100)
         )
 
-        # If no onnx model was given, generate a new one.
-        if onnx_graph is None:
-            numpy_module = NumpyModule(
-                self.torch_model,
-                dummy_input=torch.from_numpy(inputset[0]),
-            )
-
-            onnx_graph = numpy_module.onnx_model
-
-        # Get the proxy function and parameter mappings for initializing the compiler
-        self.onnx_graph = onnx_graph
-        numpy_filter = get_equivalent_numpy_forward(onnx_graph)
+        # Convert the Torch module to a Numpy module
+        numpy_module = NumpyModule(
+            self.torch_model,
+            dummy_input=torch.from_numpy(inputset[0]),
+        )
 
-        numpy_filter_proxy, parameters_mapping = generate_proxy_function(numpy_filter, ["inputs"])
+        # Get the proxy function and parameter mappings used for initializing the compiler
+        # This is done in order to be able to provide any modules with arbitrary numbers of 
+        # encrypted arguments to Concrete Numpy's compiler
+        numpy_filter_proxy, parameters_mapping = generate_proxy_function(
+            numpy_module.numpy_forward, 
+            ["inputs"]
+        )
 
+        # Compile the filter and retrieve its FHE circuit
         compiler = Compiler(
             numpy_filter_proxy,
             {parameters_mapping["inputs"]: "encrypted"},
         )
-
-        # Compile the filter
         self.fhe_circuit = compiler.compile(inputset)
 
         return self.fhe_circuit

filters/black and white/deployment/client.zip

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fa7aa78811be0810d523a8d94a1aa27e24e40d29eced0395fd3bc743568b62b8
-size 550
+oid sha256:d715c22a7efedd7e592bc04e4e7f722e2a72de12fadcd526bc2dde6465053d51
+size 388

filters/black and white/deployment/serialized_processing.json

@@ -1 +0,0 @@
-{"filter_name": "black and white"}

filters/black and white/deployment/server.zip

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5fe0885c010b076062a9b5887d0ee5ebf07f7b879633324af1b14e58a2fefeec
+oid sha256:4477689ba8760c3ee2239f25b783cb4629df831fbee9a6fde063b047219a4507
 size 4364

filters/black and white/server.onnx

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:5699206ad340e82b59392054f70b82327090bfee4909f4a0ffbd52614affc33f
-size 336

filters/blur/deployment/client.zip

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9fafcd6cd32109e17bad3ee5945b54c64525f3db5d3e893a5618cfb765a8748e
-size 542
+oid sha256:07437fc6167a00e2fc72427254829c466600f5baacaa75abab6d023f07ebd657
+size 391

filters/blur/deployment/serialized_processing.json

@@ -1 +0,0 @@
-{"filter_name": "blur"}

filters/blur/deployment/server.zip

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:49b6c8a391e67ba424f156ef6049175e5c49b13d5b92052fddf05214741175c6
+oid sha256:94ecbd3c5a43abd6864538f92aec8634b1e29191c5f8f5925effc25ae1906213
 size 7263

filters/blur/server.onnx

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:b8fd3d313ec3a9d565a0621921768317f66e53596ad950ca2be6b1efbcf984bd
-size 532

filters/identity/deployment/client.zip

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:19a7d7831af7f4a7a55a734a12e772ec41058502138e15925e229c89fcd8b195
-size 533
+oid sha256:0245406eed38f2a0c5f0afdacd5aab3746b5c7f3542f77a1ac2962355d732dbb
+size 378

filters/identity/deployment/serialized_processing.json

@@ -1 +0,0 @@
-{"filter_name": "identity"}

filters/identity/deployment/server.zip

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d2891ffa3e35d14d40a79b533fb331d557c82b4a8fe20568aa095aa7a22164a9
+oid sha256:78d4b454bd5fe0d624bacfc9119e18c8c334f9e2db6c83e5f3a212d76a9ea99a
 size 2559

filters/identity/server.onnx

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:71a8a398ea2edac9b0dfd41232c74549d1b8c159d391a4b3d42e2b4b731da02b
-size 155

filters/inverted/deployment/client.zip

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:67169abe3c33f7c7f377cd7e3b17031dd43054432a9d1b39f5469417156b5f2d
-size 533
+oid sha256:0245406eed38f2a0c5f0afdacd5aab3746b5c7f3542f77a1ac2962355d732dbb
+size 378

filters/inverted/deployment/serialized_processing.json

@@ -1 +0,0 @@
-{"filter_name": "inverted"}

filters/inverted/deployment/server.zip

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:781488531b0049ecd05d3cc0e0eb95a9350553848bf218e726e97dce2b3ebd42
+oid sha256:d673c0980ec6987ff88b96c8c642082377de4158153a0828430717284575b629
 size 4179

filters/inverted/server.onnx

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:b2fead709dbc8f0ab1f19ff1878e0ac9ce110b2b3ced261d7a87d32e0fc58b61
-size 211

filters/ridge detection/deployment/client.zip

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:05b54c87d88297316aeb864d7292c9a4c930d486e7d0b7232bdf77e9b76a7692
-size 559
+oid sha256:2d27edd60ab1d7186a6053209211922e256e86d9ae223f7056d8932890d3c50e
+size 397

filters/ridge detection/deployment/serialized_processing.json

@@ -1 +0,0 @@
-{"filter_name": "ridge detection"}

filters/ridge detection/deployment/server.zip

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:238980dd76c8155164b84d0096d11a8cbba25c933f4335fc7369e77f2328bd26
+oid sha256:c3839b27ece3a2bd7caf37a5285674a88abf026c6f8bff550c47a84b4846f0ac
 size 5043

filters/ridge detection/server.onnx

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:42f9914e64003c33c7eceb639a001ceb4460c8226e0e380cb032741851e41c49
-size 648

filters/rotate/deployment/client.zip

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a7a3c2ae45ef9887682e3e89d4138b2ce74b8e560b858f3adc0461f98f223f3f
-size 531
+oid sha256:0245406eed38f2a0c5f0afdacd5aab3746b5c7f3542f77a1ac2962355d732dbb
+size 378

filters/rotate/deployment/serialized_processing.json

@@ -1 +0,0 @@
-{"filter_name": "rotate"}

filters/rotate/deployment/server.zip

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a92a49387f05f4548cb4910506e66a6a2fa591b8f27818934d4283c8c2981a99
+oid sha256:3abdc1b55d1c63783b2d081ff67ca47c20b02ae1fd462a6398f7a80318c8529d
 size 4431

filters/rotate/server.onnx

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:aa03ea9a684b65c29c2cc0e6ab20f6b6349f35c4bd70921d264e74298a758de1
-size 178

filters/sharpen/deployment/client.zip

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8cf53584e83a91cb975e9f078ef63e777f11453582b664f65685b3a6da89f17e
-size 550
+oid sha256:21c933a828da907162224c3d13fb1b546316fa0be838ee6ee94659a27302ae67
+size 396

filters/sharpen/deployment/serialized_processing.json

@@ -1 +0,0 @@
-{"filter_name": "sharpen"}

filters/sharpen/deployment/server.zip

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4710fbe92afdd4f8f6beff7eca302a46ee4fde5b85fe8aa7d6ab832080ae5a2e
+oid sha256:a137527248ec826be5b971f645e80be95c614f82c47225d73fc11d803df93d26
 size 7311

filters/sharpen/server.onnx

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:c7958a3c9be1b578486ec1708701340263ce3ad70b7cd3ff281230797f67de0d
-size 532

generate_dev_files.py

@@ -1,7 +1,6 @@
 "A script to generate all development files necessary for the image filtering demo."
 
 import shutil
-import onnx
 from common import AVAILABLE_FILTERS, FILTERS_PATH
 from filters import Filter
 from client_server_interface import FHEDev
@@ -17,11 +16,8 @@ for filter_name in AVAILABLE_FILTERS:
     # Compile the model on a representative inputset
     filter.compile()
 
-    # Define the directory path associated to this filter
-    filter_path = FILTERS_PATH / filter_name
-
     # Define the directory path associated to this filter's deployment files
-    deployment_path = filter_path / "deployment"
+    deployment_path = FILTERS_PATH / (filter_name + "/deployment")
 
     # Delete the deployment folder and its content if it already exists
     if deployment_path.is_dir():
@@ -31,7 +27,4 @@ for filter_name in AVAILABLE_FILTERS:
     fhe_dev_filter = FHEDev(filter, deployment_path)
     fhe_dev_filter.save()
 
-    # Save the ONNX model
-    onnx.save(filter.onnx_graph, filter_path / "server.onnx")
-
 print("Done !")