Situation: Fawkes is a brilliant research project. I personally enjoy…

…ed reading the research paper. However, numerous issues and pull requests have resulted from how difficult it is to run after initially downloading it due to poorly documented python version requirements and compatibility collapse with numerous dependencies following ~4 years of drift. Example links in the last paragraph for brevity.

Action: I rewrote the setup code to depend on pyenv and poetry allowing us to explicitly require the use of python 3.9.0 rather than having users guess at how best to execute the code. I've used stricter dependency requirements (lots of ~= in pyproject.toml dependencies) as the project is unlikely to be maintained regularly and reviving the code required a lot of dependency incompatibility navigation we should avoid for future users.

Result: A new user can build this great research project with ~10 lines! A big win in my opinion.

Some relevant pull requests and issues.
#168
#158
#186
#178

.python-version

@@ -0,0 +1 @@
+3.9.0

README.md

@@ -18,6 +18,25 @@ This code is intended only for personal privacy protection or academic research.
 Usage
 -----
 
+### Local Development (Most reliable and future-proof method of running fawkes)
+
+```commandline
+# Install pyenv before running these commands
+# Instructions to install pyenv vary per OS.
+pyenv install 3.9.0
+pyenv global 3.9.0
+pip install poetry 
+# After this point global python can be anything
+# the fawkes repos .python-version file tells pyenv to use 3.9.0 when inside the fawkes folder.
+
+pip install poetry # Install poetry
+
+pyenv install 3.9.0
+cd fawkes
+poetry install
+poetry run python .\fawkes\protection.py -d ./imgs --mode low
+```
+
 `$ fawkes`
 
 Options:
@@ -34,9 +53,6 @@ Options:
 
 `fawkes -d ./imgs --mode low`
 
-or `python3 protection.py -d ./imgs --mode low`
-
-
 ### Tips
 
 - The perturbation generation takes ~60 seconds per image on a CPU machine, and it would be much faster on a GPU

fawkes/differentiator.py

@@ -195,7 +195,7 @@ def compute_batch(self, source_imgs, target_imgs=None, retry=True):
                                     dtype=tf.float32)
 
         # make the optimizer
-        optimizer = tf.keras.optimizers.Adadelta(float(self.learning_rate))
+        optimizer = tf.keras.optimizers.legacy.Adadelta(float(self.learning_rate))
         const_numpy = np.ones(len(source_imgs)) * self.initial_const
         self.const = tf.Variable(const_numpy, dtype=np.float32)
 

fawkes/protection.py

@@ -75,7 +75,7 @@ def mode2param(self, mode):
             extractors = ["extractor_0", "extractor_2"]
 
         else:
-            raise Exception("mode must be one of 'min', 'low', 'mid', 'high'")
+            raise Exception("mode must be one of 'low', 'mid', 'high'")
         return th, max_step, lr, extractors
 
     def run_protection(self, image_paths, th=0.04, sd=1e7, lr=10, max_step=500, batch_size=1, format='png',

publish.sh

@@ -0,0 +1,11 @@
+#!/bin/bash
+
+# Publish the package
+poetry publish --build
+
+# Get the current version
+VERSION=$(poetry version -s)
+
+# Create and push a Git tag
+git tag v$VERSION
+git push origin v$VERSION

pyproject.toml

@@ -0,0 +1,48 @@
+[project]
+name = "fawkes"
+version = "0.1.0"
+description = ""
+authors = [
+    {name = "Shawn Shan", email = "[email protected]"}
+]
+license = "BSD"
+homepage = "https://github.com/Shawn-Shan/fawkes"
+keywords = ["fawkes", "privacy", "ML"]
+classifiers = [
+    "Development Status :: 3 - Alpha",
+    "License :: OSI Approved :: MIT License",
+    "Operating System :: OS Independent",
+    "Programming Language :: Python :: 3",
+    "Topic :: System :: Monitoring",
+]
+
+readme = "README.md"
+requires-python = "~=3.9"
+dependencies = [
+    "numpy (>=1.22,<2.0)",
+    "tensorflow-io-gcs-filesystem (>=0.23.1,<0.24.0)",
+    "tensorflow (==2.12)",
+    "keras (>=2.12.0,<2.13.0)",
+    "setuptools (>=75.8.0,<76.0.0)",
+    "mtcnn (>=0.1.0,<0.2.0)",
+    "pillow (>=11.1.0,<12.0.0)",
+    "bleach (>=6.2.0,<7.0.0)",
+    "pyqt5 (==5.15.2)",
+    # Despite the name amd runs this implementation fine.
+    # There's no built in poetry env variable that'll tell us if an AMD GPU is present (which would imply tensorflow-rocm could be a more efficient implementation).
+    # A script could probably be built that intelligently runs pip install under the hood but this is hard to track for a newbie.
+    "tensorflow-intel (==2.12.0)"
+
+]
+
+
+[tool.poetry.dependencies]
+python = "~3.9"
+
+[tool.poetry.scripts]
+fawkes = "fawkes:main"
+
+
+[build-system]
+requires = ["poetry-core>=2.0.0,<3.0.0"]
+build-backend = "poetry.core.masonry.api"