Added a GPU check for NVIDIA in Makefile

This fix will only work on Linux. We would need to use 'ioreg' for Mac

.gitignore

@@ -113,3 +113,4 @@ venv.bak/
 *.model
 *.csv
 *.txt
+!requirements.txt

Makefile

@@ -0,0 +1,99 @@
+###############################################################################
+# Parts of this are borrowed from the HTTPie Makefile. Thanks!
+###############################################################################
+
+REQUIREMENTS="requirements-dev.txt"
+TAG="\n\n\033[0;32m\#\#\# "
+END=" \#\#\# \033[0m\n"
+
+PIP=$(shell which pip3)
+CONDA=$(shell which conda)
+CONDA_HOME="$(HOME)/anaconda3"
+VENV="eyeballer_venv"
+
+SHELL := /bin/bash
+.ONE_SHELL:
+
+all: test
+
+# Target-specific assignments
+conda: PIP := "$(CONDA_HOME)/envs/$(VENV)/bin/pip"
+conda init: uninstall-eyeballer
+
+ifeq ($(strip $(MAKECMDGOALS)),conda)
+	@echo -e $(TAG)Conda detected at: $(CONDA)$(END)
+	conda create -yn $(VENV) | true
+
+	source $(CONDA_HOME)/etc/profile.d/conda.sh; \
+	conda activate $(VENV); \
+	conda install -y pip
+endif
+	@echo -e $(TAG)Installing dev requirements$(END)
+	$(PIP) install --upgrade -r $(REQUIREMENTS)
+
+	@echo -e $(TAG)Installing Eyeballer$(END)
+ifeq ($(shell lspci | grep NVIDIA),NVIDIA))
+	$(PIP) install --upgrade --editable .[tf_gpu]
+else
+	$(PIP) install --upgrade --editable .[tf]
+endif
+
+ifeq ($(strip $(MAKECMDGOALS)),conda)
+	@echo -e $(TAG)Activate virtualenv to run eyeballer: $(END)
+	@echo conda activate $(VENV)
+endif
+
+	@echo
+
+clean:
+	@echo -e $(TAG)Cleaning up$(END)
+	rm -rf *.egg dist build .cache .pytest_cache eyeballer.egg-info
+	find . -name '__pycache__' -delete -print -o -name '*.pyc' -delete -print
+ifneq ($(strip $(CONDA)),)
+	conda env remove -y --name eyeballer_venv | true
+endif
+	@echo
+
+
+###############################################################################
+# Testing
+###############################################################################
+
+
+test:
+	@echo -e $(TAG)Running tests on the current Python interpreter with coverage $(END)
+	python -m unittest tests
+	@echo
+
+
+# test-all is meant to test everything — even this Makefile
+test-all: uninstall-all clean init test pycodestyle
+	@echo
+
+
+pycodestyle:
+	which pycodestyle || $(PIP) install pycodestyle
+	pycodestyle
+	@echo
+
+
+###############################################################################
+# Uninstalling
+###############################################################################
+
+uninstall-eyeballer:
+	@echo using pip3 located at: $(PIP)
+	@echo -e $(TAG)Uninstalling eyeballer$(END)
+	- $(PIP) uninstall --yes eyeballer &2>/dev/null
+
+	@echo "Verifying…"
+	cd .. && ! python -m eyeballer --version &2>/dev/null
+
+	@echo "Done"
+	@echo
+
+
+uninstall-all: uninstall-eyeballer
+
+	@echo -e $(TAG)Uninstalling development requirements$(END)
+	- $(PIP) uninstall --yes -r $(REQUIREMENTS)

README.md

@@ -0,0 +1,56 @@
+# Eyeballer
+Give those screenshots of yours a quick eyeballing.
+
+Eyeballer is meant for large-scope network penetration tests where you need to find "interesting" targets from a huge set of web-based hosts. Go ahead and use your favorite screenshotting tool like normal (EyeWitness or GoWitness) and then run them through Eyeballer to tell you what's likely to contain vulnerabilities, and what isn't.
+
+## Installation
+
+To install in a python venv (Recommended):
+```
+python3 -m venv venv
+source venv/bin/activate
+make init
+```
+
+To install for Conda (automatically creates conda venv):
+
+```
+make conda
+```
+
+## Running Tests
+
+```
+make test
+```
+
+## Predicting Labels
+To eyeball some screenshots, just run the "predict" mode:
+
+```
+eyeballer --weights YOUR_WEIGHTS.h5 predict YOUR_FILE.png
+```
+
+Or for a whole directory of files:
+
+```
+eyeballer --weights YOUR_WEIGHTS.h5 predict PATH_TO/YOUR_FILES/
+```
+
+## Training
+To train a new model, run:
+```
+eyeballer train
+```
+
+You'll want a machine with a good GPU for this to run in a reasonable amount of time. Setting that up is outside the scope of this readme, however.
+
+This will output a new model file (weights.h5 by default).
+
+## Evaluation
+
+You just trained a new model, cool! Let's see how well it performs against some images it's never seen before, across a variety of metrics:
+
+```
+eyeballer --weights YOUR_WEIGHTS.h5 evaluate
+```

eyeballer/__init__.py

@@ -1 +1 @@
-from eyeballer.eyeballer import EyeballModel  # noqa: F401
+from eyeballer.model import EyeballModel  # noqa: F401

eyeballer/augmentation.py

@@ -0,0 +1,18 @@
+import numpy as np
+
+from Augmentor.Operations import Operation
+from keras.applications.mobilenet import preprocess_input
+
+
+class EyeballerAugmentation(Operation):
+    def __init__(self, probability=1):
+        Operation.__init__(self, probability)
+
+    # Class must implement the perform_operation method
+    def perform_operation(self, images):
+        return_list = []
+        for image in images:
+            image_array = np.array(image).astype('uint8')
+            image_array = preprocess_input(image_array)
+            return_list.append(image_array)
+        return return_list

eyeballer/eyeballer.py

@@ -1,7 +1,10 @@
 #!/usr/bin/env python3
-import click
-from model import EyeballModel, DATA_LABELS
 
+import click
+import csv
+
+from eyeballer.model import EyeballModel, DATA_LABELS
+from eyeballer.visualization import HeatMap
 
 
 @click.group(invoke_without_command=True)
@@ -32,22 +35,41 @@ def train(ctx, graphs, batchsize, epochs):
 
 
 @cli.command()
-@click.argument('screenshot', nargs=-1)  # Nargs=-1 means you can pass as many in as you want
+@click.argument('screenshot')
+@click.option('--heatmap', default=False, is_flag=True, help="Create a heatmap graphfor the prediction")
 @click.pass_context
-def predict(ctx, screenshot):
+def predict(ctx, screenshot, heatmap):
     model = EyeballModel(**ctx.obj['model_kwargs'])
-    for fn in screenshot:
-        model.predict(fn)
+    results = model.predict(screenshot)
+
+    if heatmap:
+        # Generate a heatmap
+        HeatMap(screenshot, model).generate()
+
+    if not results:
+        print("Error: Input file does not exist")
+    if len(results) == 1:
+        print(results)
+    else:
+        with open("results.csv", "w", newline="") as csvfile:
+            fieldnames = ["filename", "custom404", "login", "homepage", "oldlooking"]
+            labelwriter = csv.DictWriter(csvfile, fieldnames=fieldnames)
+            labelwriter.writeheader()
+            labelwriter.writerows(results)
+        print("Output written to results.csv")
 
 
 def pretty_print_evaluation(results):
     """Print a human-readable summary of the evaluation"""
-    # We use 4.0% to handle all the way from 0% (2chars) to 100% (4chars)
+    # We use 4.2% to handle all the way from "  0.00%" (7chars) to "100.00%" (7chars)
     for label in DATA_LABELS:
-        print(f"{label} Precision Score: {results[label]['Precision']:4.0%}")
-        print(f"{label} Precision Score: {results[label]['Recall']:4.0%}")
-    print(f"Overall Binary Accuracy: {results['total_binary_accuracy']:4.0%}")
-    print(f"All or nothing accuracy: {results['all_or_nothing_accuracy']:4.0%}")
+        print(f"{label} Precision Score: {results[label]['precision']:4.2%}")
+        print(f"{label} Recall Score: {results[label]['recall']:4.2%}")
+    print(f"'None of the above' Precision: {results['none_of_the_above_precision']:4.2%}")
+    print(f"'None of the above' Recall: {results['none_of_the_above_recall']:4.2%}")
+    print(f"All or nothing Accuracy: {results['all_or_nothing_accuracy']:4.2%}")
+    print(f"Overall Binary Accuracy: {results['total_binary_accuracy']:4.2%}")
+    print(f"Top 10 worst predictions: {results['top_10_worst'][1]}")
 
 
 @cli.command()

eyeballer/model.py

@@ -1,10 +1,15 @@
 import os
 import random
-import time
+import sys
 
+# Prevent Tkinter Dependency
+import matplotlib
+matplotlib.use('agg')  # noqa: E402
 import matplotlib.pyplot as plt
+
 import numpy as np
 import pandas as pd
+import Augmentor
 
 from keras import regularizers
 from keras.applications import MobileNet
@@ -16,9 +21,23 @@ from keras.models import Model
 from keras.optimizers import Adam
 from keras.preprocessing import image
 from keras.preprocessing.image import ImageDataGenerator
-from sklearn.metrics import precision_score, recall_score
+from sklearn.metrics import classification_report, accuracy_score, hamming_loss
+from eyeballer.augmentation import EyeballerAugmentation
+from keras.callbacks import Callback
 
-DATA_LABELS = ["custom404", "login", "homepage"]
+DATA_LABELS = ["custom404", "login", "homepage", "oldlooking"]
+
+
+class PredictionManager(Callback):
+    def __init__(self):
+        super(PredictionManager, self).__init__()
+        self.predictions = []
+
+    def getPredictions(self):
+        return np.array(self.predictions)
+
+    def on_predict_batch_end(self, batch, logs=None):
+        self.predictions.append(logs['outputs'][0].tolist()[0])
 
 
 class EyeballModel:
@@ -26,12 +45,13 @@ class EyeballModel:
 
     Contains high-level functions for training, evaluating, and predicting.
     """
-    checkpoint_file = "weights-snapshot.h5"
+    graphs_directory = "graphs/"
+    checkpoint_file = "weights.h5"
     image_dir = "images_224x224/"
     image_width, image_height = 224, 224
     input_shape = (image_width, image_height, 3)
 
-    def __init__(self, print_summary=False, weights_file="weights.h5", seed=None):
+    def __init__(self, weights_file, print_summary=False, seed=None):
         """Constructor for model class.
 
         Keyword arguments:
@@ -47,11 +67,11 @@ class EyeballModel:
         x = Dropout(0.5)(x)
         x = Dense(128, activation="relu", name="HiddenLayer2", kernel_regularizer=regularizers.l2(0.01))(x)
         x = Dropout(0.2)(x)
-        output_layer = Dense(3, activation="sigmoid", name="OutputLayer")(x)
+        output_layer = Dense(len(DATA_LABELS), activation="sigmoid", name="OutputLayer")(x)
         self.model = Model(inputs=base_model.input, outputs=output_layer)
 
         # for layer in base_model.layers:
-        #    layer.trainable = False
+        #     layer.trainable = False
 
         adam = Adam(lr=0.0005)
         self.model.compile(optimizer=adam, loss="binary_crossentropy", metrics=["accuracy"])
@@ -76,10 +96,30 @@ class EyeballModel:
         self.training_labels = self.training_labels.sample(frac=1)
 
         if weights_file is not None and os.path.isfile(weights_file):
-            self.model.load_weights(weights_file)
+            try:
+                self.model.load_weights(weights_file)
+            except OSError:
+                print(f"ERROR: Unable to open weights file '{weights_file}'")
+                sys.exit(-1)
             print("Loaded model from file.")
         else:
-            print("No model loaded from file")
+            if weights_file is not None:
+                raise FileNotFoundError
+            print("WARN: No model loaded from file. Generating random model")
+
+        # Data augmentation
+        augmentor = Augmentor.Pipeline()
+        augmentor.set_seed(self.seed)
+        augmentor.zoom(probability=0.75, min_factor=0.8, max_factor=1.2)
+        augmentor.random_color(probability=0.75, min_factor=0.5, max_factor=1.0)
+        augmentor.random_contrast(probability=0.75, min_factor=0.8, max_factor=1.0)
+        augmentor.random_brightness(probability=0.75, min_factor=0.8, max_factor=1.2)
+        augmentor.random_erasing(probability=0.75, rectangle_area=0.15)
+
+        # Finalizes the augmentation with a custom operation to prepare the image for the specific pretrained model we're using
+        training_augmentation = EyeballerAugmentation()
+        augmentor.add_operation(training_augmentation)
+        self.preprocess_training_function = augmentor.keras_preprocess_func()
 
     def train(self, epochs=20, batch_size=32, print_graphs=False):
         """Train the model, making a new weights file at each successfull checkpoint. You'll probably need a GPU for this to realistically run.
@@ -91,15 +131,10 @@ class EyeballModel:
         """
         print("Training with seed: " + str(self.seed))
 
-        # Data augmentation
         data_generator = ImageDataGenerator(
-            preprocessing_function=preprocess_input,
-            shear_range=0.0,
-            zoom_range=0.2,
-            brightness_range=(0.7, 1.3),
-            samplewise_center=True,
+            preprocessing_function=self.preprocess_training_function,
             validation_split=0.2,
-            horizontal_flip=False)
+            samplewise_center=True)
 
         training_generator = data_generator.flow_from_dataframe(
             self.training_labels,
@@ -138,6 +173,8 @@ class EyeballModel:
             verbose=1)
 
         if print_graphs:
+            if not os.path.exists(self.graphs_directory):
+                os.makedirs(self.graphs_directory)
             # Plot training & validation accuracy values
             plt.plot(history.history['acc'])
             plt.plot(history.history['val_acc'])
@@ -145,7 +182,7 @@ class EyeballModel:
             plt.ylabel('Accuracy')
             plt.xlabel('Epoch')
             plt.legend(['Train', 'Validation'], loc='upper left')
-            plt.savefig("accuracy.png")
+            plt.savefig(self.graphs_directory + "accuracy.png")
             plt.clf()
             plt.cla()
             plt.close()
@@ -157,28 +194,60 @@ class EyeballModel:
             plt.ylabel('Loss')
             plt.xlabel('Epoch')
             plt.legend(['Train', 'Validation'], loc='upper left')
-            plt.savefig("loss.png")
+            plt.savefig(self.graphs_directory + "loss.png")
 
-    def predict(self, filename):
+    def predict_on_array(self, image):
+        """Predict the labels for a single screenshot
+
+        Keyword arguments:
+        image -- The numpy array of the image to classify
+        """
+        img = np.expand_dims(image, axis=0)
+        img = preprocess_input(img)
+
+        prediction = self.model.predict(img, batch_size=1)
+        result = dict()
+        result["filename"] = "custom-image"
+        result["custom404"] = prediction[0][0]
+        result["login"] = prediction[0][1]
+        result["homepage"] = prediction[0][2]
+        result["oldlooking"] = prediction[0][3]
+        return result
+
+    def predict(self, path):
         """Predict the labels for a single file
 
         Keyword arguments:
-        filename -- The path to the file that we'll be evaluating
+        path -- The path to the file that we'll be evaluating
         """
-        # Load the image into memory
-        img = image.load_img(filename, target_size=(self.image_width, self.image_height))
-        img = image.img_to_array(img)
-        img = np.expand_dims(img, axis=0)
-        img = preprocess_input(img)
+        # Is this a single file, or a directory?
+        screenshots = []
+        if os.path.isfile(path):
+            screenshots = [path]
+        elif os.path.isdir(path):
+            screenshots = os.listdir(path)
+            screenshots = [os.path.join(path, s) for s in screenshots]
+        else:
+            raise FileNotFoundError
 
-        time_start = time.time()
-        prediction = self.model.predict(img, batch_size=1)
-        time_end = time.time()
-        print("Predictions:")
-        print("\tCustom 404: " + str(round(prediction[0][0] * 100, 2)) + "%")
-        print("\tLogin Page: " + str(round(prediction[0][1] * 100, 2)) + "%")
-        print("\tHome Page: " + str(round(prediction[0][2] * 100, 2)) + "%")
-        print("Prediction Took (seconds): ", time_end - time_start)
+        results = []
+        for screenshot in screenshots:
+            # Load the image into memory
+            img = image.load_img(screenshot, target_size=(self.image_width, self.image_height))
+            img = image.img_to_array(img)
+            img = np.expand_dims(img, axis=0)
+            img = preprocess_input(img)
+
+            prediction = self.model.predict(img, batch_size=1)
+            result = dict()
+            result["filename"] = screenshot
+            result["custom404"] = prediction[0][0]
+            result["login"] = prediction[0][1]
+            result["homepage"] = prediction[0][2]
+            result["oldlooking"] = prediction[0][3]
+            results.append(result)
+
+        return results
 
     def evaluate(self, threshold=0.5):
         """Evaluate performance against the persistent evaluation data set
@@ -203,13 +272,9 @@ class EyeballModel:
             print("Using validation set...")
             # Data augmentation
             data_generator = ImageDataGenerator(
-                preprocessing_function=preprocess_input,
-                shear_range=0.0,
-                zoom_range=0.2,
-                brightness_range=(0.7, 1.3),
+                preprocessing_function=self.preprocess_training_function,
                 samplewise_center=True,
-                validation_split=0.2,
-                horizontal_flip=False)
+                validation_split=0.2)
             evaluation_generator = data_generator.flow_from_dataframe(
                 self.training_labels,
                 directory=self.image_dir,
@@ -224,39 +289,107 @@ class EyeballModel:
         else:
             print("Using evaluation set...")
 
-        predictions = self.model.predict_generator(evaluation_generator, verbose=1, steps=len(evaluation_generator))
+        prediction_manager = PredictionManager()
+
+        try:
+            self.model.predict_generator(
+                evaluation_generator,
+                callbacks=[prediction_manager],
+                verbose=1,
+                steps=len(evaluation_generator))
+        except KeyboardInterrupt:
+            print('\r\nJob interrupted (Ctrl+C): Outputting completed results')
+
+        predictions = prediction_manager.getPredictions()
+
+        # Truncates evaluation generator if test was interrupted
+        evaluation_generator._data = evaluation_generator.data[:predictions.shape[0]]
+
+        self._save_prediction_histograms(predictions)
         predictions = predictions > threshold
-
-        all_or_nothing_success = 0
-
-        correct_per_category_counts = [0, 0, 0]
-        total_correct_count = 0
-
-        labels = evaluation_generator.data.tolist()
-
-        # Zip them together, so we can evaluate easier
-        for row in zip(labels, predictions.tolist()):
-            # All or nothing count
-            if row[0] == row[1]:
-                all_or_nothing_success += 1
-
-            # Per-category count
-            for i in range(len(row[0])):
-                if row[0][i] == row[1][i]:
-                    correct_per_category_counts[i] += 1
-                    total_correct_count += 1
-
-        stats = {label: dict() for label in DATA_LABELS}
-        for index, scores in enumerate(stats.values()):
-            data_slice = np.index_exp[:, index:index+1]  # variable is verbosely named to avoid shadowing np.slice
-            labels_column = self.get_data_column(data_slice, labels)
-            predictions_column = self.get_data_column(data_slice, predictions)
-            scores['Precision'] = precision_score(labels_column, predictions_column)
-            scores['Recall'] = recall_score(labels_column, predictions_column)
-
-        stats["total_binary_accuracy"] = total_correct_count / (len(evaluation_generator) * len(DATA_LABELS))
-        stats["all_or_nothing_accuracy"] = all_or_nothing_success / len(evaluation_generator)
+        ground_truth = evaluation_generator.data
+        stats = classification_report(ground_truth, predictions, target_names=DATA_LABELS, output_dict=True)
+        stats["total_binary_accuracy"] = 1 - hamming_loss(ground_truth, predictions)
+        stats["all_or_nothing_accuracy"] = accuracy_score(ground_truth, predictions)
+        stats["top_10_best"] = self._top_images(evaluation_generator, predictions, best=True)
+        stats["top_10_worst"] = self._top_images(evaluation_generator, predictions, best=False)
+        stats["none_of_the_above_recall"] = self._none_of_the_above_recall(ground_truth, predictions)
+        stats["none_of_the_above_precision"] = self._none_of_the_above_precision(ground_truth, predictions)
         return stats
 
-    def get_data_column(self, data_slice, data):
-        return np.reshape(np.array(data)[data_slice], len(data)).tolist()
+    def _none_of_the_above_recall(self, labels, predictions):
+        """Returns the recall score for the 'none of the above' images.
+        That means, all the images that don't have a category.
+        """
+        total_count = 0
+        correct_count = 0
+        for item in zip(labels.astype(int), predictions.astype(int)):
+            # Is this a none of the above?
+            if not item[0].any():
+                total_count += 1
+                if not item[1].any():
+                    correct_count += 1
+        return correct_count / total_count
+
+    def _none_of_the_above_precision(self, labels, predictions):
+        """Returns the precision score for the 'none of the above' images.
+        That means, all the images that don't have a category.
+        """
+        total_count = 0
+        correct_count = 0
+        for item in zip(labels.astype(int), predictions.astype(int)):
+            # Is this a none of the above prediction?
+            if not item[1].any():
+                total_count += 1
+                if not item[0].any():
+                    correct_count += 1
+        return correct_count / total_count
+
+    def _top_images(self, labels, predictions, top_k=10, best=False):
+        """Collect top-k best or top-k worst predicted images
+
+        Keyword arguments:
+        labels -- The keras generator that contains the filenames and data labels
+        predictions -- The numpy array of predictions
+        top_k -- Top k elements
+        best -- True/False. Calculate either the best or worst images
+
+        :Return -- Tuple of top-k indicies and top-k filenames
+        """
+        true_labels = np.array(labels.data).astype(float)
+        predictions = predictions.astype(float)
+
+        differences = np.abs(predictions - true_labels).sum(axis=1)
+        indicies = np.argsort(differences, axis=0)
+
+        top_file_list = []
+
+        if not best:
+            # Reverse numpy array
+            indicies = np.flipud(indicies)
+
+        for i in indicies[:top_k]:
+            top_file_list.append(labels.filenames[i])
+
+        return indicies[:top_k], top_file_list
+
+    def _save_prediction_histograms(self, predictions, buckets=50):
+        """Saves a series of histogram screenshots
+
+            Keyword arguments:
+            predictions -- The numpy array of predicted labels, ranging from 0->1.
+            buckets -- The number of buckets to divide the data into. Default: 50
+            :Returns -- Nothing"
+        """
+        figure, axes = plt.subplots(nrows=len(DATA_LABELS))
+        for i, label in enumerate(DATA_LABELS):
+            axes[i].hist(predictions[:, i], buckets, alpha=.75)
+            axes[i].set_xlabel("Prediction")
+            axes[i].set_ylabel("Counts of Predictions")
+            axes[i].set_title(label)
+            axes[i].grid(True)
+        figure.set_size_inches(5, 3*len(DATA_LABELS))
+        figure.tight_layout()
+        if not os.path.exists(self.graphs_directory):
+            os.makedirs(self.graphs_directory)
+        plt.savefig(self.graphs_directory + "label_histograms.png")

eyeballer/visualization.py

@@ -0,0 +1,127 @@
+import numpy as np
+import math
+from keras.preprocessing import image
+from copy import deepcopy
+from matplotlib import pyplot as plt
+from eyeballer.model import DATA_LABELS
+
+
+class HeatMap():
+    def __init__(self, screenshot_path, model, boxsize=28, step=7):
+        self.model = model
+
+        img = image.load_img(screenshot_path, target_size=(self.model.image_width, self.model.image_height))
+        img = image.img_to_array(img)
+
+        self.screenshot = img
+        self.boxsize = boxsize
+        self.step = step
+        self.x = 0
+        self.y = 0
+        self.screenshot_path = screenshot_path
+        self.gamma = 3
+
+    def generate(self, output_file="heatmap.png"):
+        """ Make a single heatmap image and return it """
+        heatmaps = []
+        labels = []
+        results = self.model.predict(self.screenshot_path)
+        results = results[0]
+        for label in DATA_LABELS:
+            boxsize = self.boxsize
+            # Ignore this label if it didn't predict positively (true label)
+            if results[label] > 0.5:
+                worst_score = 1
+                while worst_score > 0.5:
+                    heatmap, worst_score = self._get_heatmap(label, boxsize)
+                    if worst_score > 0.5:
+                        boxsize += 28
+                        print(f"Didn't get a good image for {label}. Trying again with a bigger boxsize: {boxsize}")
+                    else:
+                        heatmaps.append(heatmap)
+                        labels.append(label)
+
+        plt.figure()
+        screenshot_image = image.load_img(self.screenshot_path, target_size=(self.model.image_width, self.model.image_width))
+        _, subplots = plt.subplots(1, len(heatmaps))
+        # This is a little janky, but matplotlib returns a list above if there's multiple subplots, and just a single subplot if there's only one
+        #   it would have been easier if it was always a list. But alas
+        if not heatmaps:
+            print("No heatmap made. The image did not have a True classification for any label")
+            return
+        if len(heatmaps) > 1:
+            for i, heatmap in enumerate(heatmaps):
+                subplots[i].imshow(screenshot_image, cmap='binary', interpolation='none')
+                subplots[i].imshow(heatmap, alpha=0.5, interpolation='none')
+                subplots[i].set_title(labels[i])
+        else:
+            subplots.imshow(screenshot_image, cmap='binary', interpolation='none')
+            subplots.imshow(heatmap, alpha=0.5, interpolation='none')
+            subplots.set_title(labels[0])
+        plt.savefig(output_file)
+        print(f"Heatmap image written to: {output_file}")
+
+    def _get_heatmap(self, label, boxsize):
+        worst_score = 1
+        self.x, self.y = 0, 0
+
+        heatmap = np.ones((self.model.image_width, self.model.image_height))
+        heatmap *= 255
+        while True:
+            # Occlude an image
+            occluded_image, x, y = self._occlude(boxsize)
+            if occluded_image is not None:
+                # Score the new occluded image
+                results = self.model.predict_on_array(occluded_image)
+                score = results[label]  # TODO
+                worst_score = min(worst_score, score)
+                # Scale the score up to pixel values
+                score = math.floor(score * 255)
+                # Anneal the scores as you go outward into the occlusion block
+                new_scores = np.ones((boxsize, boxsize))
+                new_scores *= score
+                new_scores = self._gamma_anneal(new_scores)
+
+                # For each pixel, update only where the new score is lower
+                original_scores = heatmap[x: x+boxsize, y: y+boxsize]
+
+                target_area = np.where(new_scores < original_scores, new_scores, original_scores)
+                heatmap[x: x+boxsize, y: y+boxsize] = target_area
+            else:
+                break
+
+        return heatmap, worst_score
+
+    def _gamma_anneal(self, occlusion_area):
+        """ Get the annealed score based on centeredness of the occlusion area """
+        width, height = occlusion_area.shape
+
+        center_x = (width-1) / 2
+        center_y = (height-1) / 2
+
+        annealed_area = np.zeros(occlusion_area.shape)
+
+        for (x, y), score in np.ndenumerate(occlusion_area):
+            distance = math.sqrt((center_x - x)**2 + (center_y - y)**2)
+            annealed_area[x, y] = min(255, score + (self.gamma * distance))
+
+        return annealed_area
+
+    def _occlude(self, boxsize):
+        ''' Return a single occluded image and its location '''
+        if self.x + boxsize > self.screenshot.shape[0]:
+            return None, None, None
+
+        retImg = np.copy(self.screenshot)
+        retImg[self.x:self.x+boxsize, self.y:self.y+boxsize] = 0.0
+
+        old_i = deepcopy(self.x)
+        old_j = deepcopy(self.y)
+
+        # update indices
+        self.y = self.y + self.step
+        if self.y+boxsize > self.screenshot.shape[1]:  # reached end
+            self.y = 0  # reset j
+            self.x = self.x + self.step  # go to next row
+
+        return retImg, old_i, old_j

setup.py

@@ -6,17 +6,28 @@ setup(
     packages=find_packages(exclude=["tests"]),
     url='https://github.com/BishopFox/eyeballer',
     license='GNU General Public License v3.0',
-    author='Daniel Petro',
+    author='Bishop Fox',
     author_email='',
     description='',
+    entry_points={
+        'console_scripts': [
+            'eyeballer = eyeballer.eyeballer:cli'
+        ]
+    },
     install_requires=[
-        'keras',
-        'sklearn',
+        'Augmentor',
+        'click',
+        'matplotlib',
         'numpy',
         'pandas',
-        'matplotlib',
-        'pillow', 'click'
+        'pillow',
+        'sklearn',
+        'keras @ git+https://github.com/keras-team/keras',
+    ],
+    extras_require={
+    'tf': ['tensorflow'],
+    'tf_gpu': ['tensorflow-gpu'],
+    },
+    dependency_links=[
     ],
-    test_requires=[],
-    python_requires='>=3.6',    # We use some python 3.6+ only features, like ordered dicts by default and f-strings
 )

tests/EvaluateTest.py

@@ -0,0 +1 @@
+#TODO

tests/PredictTest.py

@@ -0,0 +1,84 @@
+import errno
+import os
+import unittest
+
+from eyeballer.model import EyeballModel
+
+
+class PredictTest(unittest.TestCase):
+    def setUp(self):
+        weights_file = "latest_weights.h5"
+        if not os.path.isfile(weights_file):
+            print("Error: Symlink the latest weights file to 'latest_weights.h5'")
+            raise FileNotFoundError(
+                errno.ENOENT,
+                os.strerror(errno.ENOENT),
+                weights_file)
+
+        model_kwargs = {
+            "weights_file": weights_file,
+            "print_summary": False,
+            "seed": None
+        }
+        self.model = EyeballModel(**model_kwargs)
+
+    def test_different_seed_predict(self):
+        model_kwargs = {
+            "weights_file": None,
+            "print_summary": False,
+            "seed": 12345678
+        }
+        same_seed_model = EyeballModel(**model_kwargs)
+
+        screenshot = "tests/data/test.png"
+
+        results_one = self.model.predict(screenshot)
+        results_two = same_seed_model.predict(screenshot)
+
+        self.assertNotEqual(results_one, results_two)
+
+    def test_same_seed_predict(self):
+        model_kwargs = {
+            "weights_file": None,
+            "print_summary": False,
+            "seed": 12345678
+        }
+        same_seed_model = EyeballModel(**model_kwargs)
+
+        screenshot = "tests/data/test.png"
+
+        results_one = same_seed_model.predict(screenshot)
+        results_two = same_seed_model.predict(screenshot)
+
+        self.assertEquals(results_one, results_two)
+
+    def get_prediction_by_name(self, results):
+        prediction = None
+        max_val = 0.0
+
+        print(results)
+        for k, v in results[0].items():
+            if not isinstance(v, str) and v > max_val:
+                prediction = k
+                max_val = v
+        return prediction
+
+    def test_predict_custom404(self):
+        screenshot = "tests/data/custom404.png"
+        results = self.model.predict(screenshot)
+        self.assertEqual(self.get_prediction_by_name(results), "custom404")
+
+    def test_predict_login(self):
+        screenshot = "tests/data/login.png"
+        results = self.model.predict(screenshot)
+        self.assertEqual(self.get_prediction_by_name(results), "login")
+
+    def test_predict_homepage(self):
+        screenshot = "tests/data/homepage.png"
+        results = self.model.predict(screenshot)
+        self.assertEqual(self.get_prediction_by_name(results), "homepage")
+
+    def test_predict_oldlooking(self):
+        screenshot = "tests/data/oldlooking.png"
+        results = self.model.predict(screenshot)
+        self.assertEqual(self.get_prediction_by_name(results), "oldlooking")

tests/TrainTest.py

@@ -0,0 +1 @@
+#TODO

tests/__init__.py

@@ -0,0 +1 @@
+from tests.PredictTest import *

utils/labelbox_to_labels.py

@@ -0,0 +1,37 @@
+#!/usr/bin/env python3
+
+import csv
+import random
+import json
+
+with open("newlabels.csv", "w", newline="") as csvfile:
+    # Open the old labels file
+    with open("labelbox.csv", newline="") as oldfile:
+        # Get the header labels
+        csvreader = csv.DictReader(oldfile)
+        fieldnames = next(csvreader)
+        labelwriter = csv.DictWriter(csvfile, fieldnames=["filename", "login", "custom404", "homepage", "oldlooking", "evaluation"])
+        labelwriter.writeheader()
+
+        # Loop through the file
+        rows = []
+        for row in csvreader:
+            filename = row["External ID"]
+            print(row["Label"])
+            labelstring = row["Label"]
+            if row["Label"] == "Skip":
+                labelstring = '{"imageclassification":[]}'
+            labels = json.loads(labelstring)
+
+            newrow = dict()
+            newrow["filename"] = filename
+            newrow["oldlooking"] = True
+            newrow["login"] = "loginpage" in labels["imageclassification"]
+            newrow["homepage"] = "homepage" in labels["imageclassification"]
+            newrow["custom404"] = "custom404"in labels["imageclassification"]
+
+            newrow["evaluation"] = random.random() > 0.8
+
+            rows.append(newrow)
+        labelwriter.writerows(rows)
+print("Made new labels file: newlabels.csv")