Merge pull request tensorflow#4135 from vincentvanhoucke/master

vincentvanhoucke · web-flow · commit 3bdc47736aac · 2018-04-30T19:27:30.000-07:00
Add marco model: automating the evaluation of crystallization experiments
diff --git a/CODEOWNERS b/CODEOWNERS
@@ -22,6 +22,7 @@
 /research/lexnet_nc/ @vered1986 @waterson
 /research/lfads/ @jazcollins @susillo
 /research/lm_1b/ @oriolvinyals @panyx0718
+/research/marco/ @vincentvanhoucke
 /research/maskgan/ @a-dai
 /research/namignizer/ @knathanieltucker
 /research/neural_gpu/ @lukaszkaiser
diff --git a/research/README.md b/research/README.md
@@ -55,6 +55,7 @@ installation](https://www.tensorflow.org/install).
 -   [pcl_rl](pcl_rl): code for several reinforcement learning algorithms,
     including Path Consistency Learning.
 -   [ptn](ptn): perspective transformer nets for 3D object reconstruction.
+-   [marco](marco): automating the evaluation of crystallization experiments.
 -   [qa_kg](qa_kg): module networks for question answering on knowledge graphs.
 -   [real_nvp](real_nvp): density estimation using real-valued non-volume
     preserving (real NVP) transformations.
diff --git a/research/marco/README.md b/research/marco/README.md
@@ -0,0 +1,78 @@
+Automating the Evaluation of Crystallization Experiments
+========================================================
+
+This is a pretrained model described in the paper:
+
+[Classification of crystallization outcomes using deep convolutional neural networks](https://arxiv.org/abs/1803.10342).
+
+This model takes images of crystallization experiments as an input:
+
+<img src="https://storage.googleapis.com/marco-168219-model/002s_C6_ImagerDefaults_9.jpg" alt="crystal sample" width="320" height="240" />
+
+It classifies it as belonging to one of four categories: crystals, precipitate, clear, or 'others'.
+
+The model is a variant of [Inception-v3](https://arxiv.org/abs/1512.00567) trained on data from the [MARCO](http://marco.ccr.buffalo.edu) repository.
+
+Model
+-----
+
+The model can be downloaded from:
+
+https://storage.googleapis.com/marco-168219-model/savedmodel.zip
+
+Example
+-------
+
+1. Install TensorFlow and the [Google Cloud SDK](https://cloud.google.com/sdk/gcloud/).
+
+2. Download and unzip the model:
+
+ ```bash
+ unzip savedmodel.zip
+ ```
+
+3. A sample image can be downloaded from:
+
+ https://storage.googleapis.com/marco-168219-model/002s_C6_ImagerDefaults_9.jpg
+
+ Convert your image into a JSON request using:
+
+ ```bash
+ python jpeg2json.py 002s_C6_ImagerDefaults_9.jpg > request.json
+ ```
+
+4. To issue a prediction, run:
+
+ ```bash
+ gcloud ml-engine local predict --model-dir=savedmodel --json-instances=request.json
+ ```
+
+The request should return normalized scores for each class:
+
+<pre>
+CLASSES                                            SCORES
+[u'Crystals', u'Other', u'Precipitate', u'Clear']  [0.926338255405426, 0.026199858635663986, 0.026074528694152832, 0.021387407556176186]
+</pre>
+
+CloudML Endpoint
+----------------
+
+The model can also be accessed on [Google CloudML](https://cloud.google.com/ml-engine/) by issuing:
+
+```bash
+gcloud ml-engine predict --model marco_168219_model --json-instances request.json
+```
+
+Ask the author for access privileges to the CloudML instance.
+
+Note
+----
+
+`002s_C6_ImagerDefaults_9.jpg` is a sample from the
+[MARCO](http://marco.ccr.buffalo.edu) repository, contributed to the dataset under the [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/) license.
+
+Author
+------
+
+[Vincent Vanhoucke](mailto:vanhoucke@google.com) (github: vincentvanhoucke)
+                                                                                
diff --git a/research/marco/jpeg2json.py b/research/marco/jpeg2json.py
@@ -0,0 +1,35 @@
+#!/usr/bin/python
+# Copyright 2018 The TensorFlow Authors. All Rights Reserved.
+#
+# 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.
+# ==============================================================================
+"""jpeg2json.py: Converts a JPEG image into a json request to CloudML.
+
+Usage:
+python jpeg2json.py 002s_C6_ImagerDefaults_9.jpg > request.json
+
+See:
+https://cloud.google.com/ml-engine/docs/concepts/prediction-overview#online_prediction_input_data
+"""
+
+import base64
+import sys
+
+
+def to_json(data):
+  return '{"image_bytes":{"b64": "%s"}}' % base64.b64encode(data)
+
+
+if __name__ == '__main__':
+  file = open(sys.argv[1]) if len(sys.argv) > 1 else sys.stdin
+  print(to_json(file.read()))
diff --git a/research/marco/request.json b/research/marco/request.json
