Add instructions to download dataset and relevant GLoVe embeddings.

PiperOrigin-RevId: 282954546

scratchgan/README.md

@@ -39,7 +39,13 @@ The data contains:
 
 ## Running
 
-Place the data files in the directory specified by `data_dir` flag.
+Download the data and place it in the directory specified by `data_dir` flag:
+
+    mkdir -p /tmp/emnlp2017
+    curl https://storage.googleapis.com/deepmind-scratchgan-data/train.json --output /tmp/emnlp2017/train.json
+    curl https://storage.googleapis.com/deepmind-scratchgan-data/valid.json --output /tmp/emnlp2017/valid.json
+    curl https://storage.googleapis.com/deepmind-scratchgan-data/test.json --output /tmp/emnlp2017/test.json
+    curl https://storage.googleapis.com/deepmind-scratchgan-data/glove_emnlp2017.txt --output /tmp/emnlp2017/glove_emnlp2017.txt
 
 Create and activate a virtual environment if needed:
 

unrestricted_advx/README.md

@@ -0,0 +1,33 @@
+# Unrestricted Adversarial Challenge
+
+This is a submission for the unrestricted adversarial challenge: Phase I. The
+entry is uses a pretrained ImageNet model with Local Linearity Regularizer, then
+adversarially trained using birds-or-bicycles dataset (train and extras) as
+provided by the challenge.
+
+> Tom B. Brown et al
+  *Unrestricted Adversarial Examples*. [\[arXiv\]](https://arxiv.org/pdf/1809.08352).
+> Chongli Qin et al
+  *Adversarial Robustness through Local Linearization*. NEURIPS 2019. [\[arXiv\]](https://arxiv.org/abs/1907.02610)
+
+
+## Contents
+
+The code contains:
+
+- a main file (`main.py`) for our submission.
+
+## Running
+
+Install requirements please follow instructions on
+[Unrestricted Adversarial Challenge.](https://github.com/google/unrestricted-adversarial-examples/blob/master/warmup.md)
+
+You can do this by running:
+
+    ./unrestricted_advx/install_dependencies.sh
+
+You can run the main script by:
+
+    ./unrestricted_advx/run.sh
+
+

unrestricted_advx/install_dependencies.sh

@@ -0,0 +1,25 @@
+#!/bin/sh
+# Copyright 2019 Deepmind Technologies Limited.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+# Usage:
+# user@host:/path/to/deepmind_research$ unrestricted_advx/run.sh
+
+# Sets up virtual environment, install dependencies, and runs evaluation script
+python3 -m venv unrestricted_venv
+source unrestricted_venv/bin/activate
+pip install -r unrestricted_advx/requirements.txt
+git clone git@github.com:google/unrestricted-adversarial-examples.git
+pip install -e unrestricted-adversarial-examples/bird-or-bicycle
+pip install -e unrestricted-adversarial-examples/unrestricted-advex

unrestricted_advx/main.py

@@ -0,0 +1,58 @@
+# Copyright 2019 Deepmind Technologies Limited.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Submission to Unrestricted Adversarial Challenge."""
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+import tensorflow as tf
+import tensorflow_hub as hub
+from unrestricted_advex import eval_kit
+
+
+def _preprocess_image(image):
+  image = tf.image.central_crop(image, central_fraction=0.875)
+  image = tf.image.resize_bilinear(image, [224, 224], align_corners=False)
+  return image
+
+
+def test_preprocess(image):
+  image = _preprocess_image(image)
+  image = tf.subtract(image, 0.5)
+  image = tf.multiply(image, 2.0)
+  return image
+
+
+def main():
+  g = tf.Graph()
+  with g.as_default():
+    input_tensor = tf.placeholder(tf.float32, (None, 224, 224, 3))
+    x_np = test_preprocess(input_tensor)
+    raw_module_1 = hub.Module(
+        "https://tfhub.dev/deepmind/llr-pretrain-adv/latents/1")
+    raw_module_2 = hub.Module(
+        "https://tfhub.dev/deepmind/llr-pretrain-adv/linear/1")
+    latents = raw_module_1(dict(inputs=x_np, decay_rate=0.1))
+    logits = raw_module_2(dict(inputs=latents))
+    logits = tf.squeeze(logits, axis=[1, 2])
+    two_class_logits = tf.concat([tf.nn.relu(-logits[:, 1:]),
+                                  tf.nn.relu(logits[:, 1:])], axis=1)
+    sess = tf.train.SingularMonitoredSession()
+    def model(x_np):
+      return sess.run(two_class_logits, feed_dict={input_tensor: x_np})
+
+    eval_kit.evaluate_bird_or_bicycle_model(model, model_name="llr_resnet")
+
+if __name__ == "__main__":
+  main()

unrestricted_advx/requirements.txt

@@ -0,0 +1,4 @@
+absl-py>=0.7.0
+numpy>=1.16.4
+tensorflow>=1.14
+tensorflow-hub>=0.5.0

unrestricted_advx/run.sh

@@ -0,0 +1,21 @@
+#!/bin/sh
+# Copyright 2019 Deepmind Technologies Limited.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+# Usage:
+# user@host:/path/to/deepmind_research$ unrestricted_advx/run.sh
+
+# Runs evaluation script
+source unrestricted_venv/bin/activate
+python3 -m unrestricted_advx.main