[MXNET-53]Image classifier for scala-infer package #10054

Roshrini · 2018-03-09T23:44:48Z

Description

https://issues.apache.org/jira/browse/MXNET-50

Depends on #9678
So, build will fail till that PR gets merged

Checklist

Essentials

Passed code style checking (make lint)
Changes are complete (i.e. I finished coding on this PR)
All changes have test coverage:
Unit tests are added for small changes to verify correctness (e.g. adding a new operator)
Nightly tests are added for complicated/long-running ones (e.g. changing distributed kvstore)
Build tests will be added for build configuration changes (e.g. adding a new build option with NCCL)
Code is well-documented:
For user-facing API changes, API doc string has been updated.
For new C++ functions in header files, their functionalities and arguments are documented.
For new examples, README.md is added to explain the what the example does, the source of the dataset, expected performance on test set and reference to the original paper if applicable
To the my best knowledge, examples are either not affected by this change, or have been fixed to be compatible with this change

Changes

Feature1, tests, (and when applicable, API doc)
Feature2, tests, (and when applicable, API doc)

Comments

If this change is a backward incompatible change, why must this change be made.
Interesting edge cases to note here

nswamy · 2018-03-10T00:17:22Z

scala-package/infer/src/main/scala/ml/dmlc/mxnet/infer/ImageClassifier.scala

+  def classifyImageBatch(inputBatch: Traversable[BufferedImage], topK: Option[Int] = None):
+  List[List[(String, Float)]] = {
+    val result = ListBuffer[List[(String, Float)]]()
+    for (image <- inputBatch) {


as discussed offline
the batch of input should be converted to an NDArray and run with ClassifywithNDArray

marcoabreu

Would you mind adding tests to verify the examples?

nswamy · 2018-03-15T18:46:30Z

scala-package/infer/src/main/scala/ml/dmlc/mxnet/infer/ImageClassifier.scala

+
+package ml.dmlc.mxnet.infer
+
+import ml.dmlc.mxnet._


Please import only the required modules

nswamy · 2018-03-15T18:48:00Z

scala-package/infer/src/main/scala/ml/dmlc/mxnet/infer/ImageClassifier.scala

+  def classifyImage(inputImage: BufferedImage,
+                        topK: Option[Int] = None): IndexedSeq[List[(String, Float)]] = {
+
+    val width = inputDescriptors(0).shape(2)


This should be based on the layout parameter of the input descriptor

nswamy · 2018-03-15T18:49:27Z

scala-package/infer/src/main/scala/ml/dmlc/mxnet/infer/ImageClassifier.scala

+    * @param newHeight rescale to new height
+    * @return Rescaled BufferedImage
+    */
+  def getScaledImage(img: BufferedImage, newWidth: Int, newHeight: Int): BufferedImage = {


why not call this reshape Image? instead of getScaledImage. Also this can be static method, so users can just call this method without having to create an object. ie.., a method on the companion object

nswamy · 2018-03-15T18:54:04Z

scala-package/infer/src/main/scala/ml/dmlc/mxnet/infer/ImageClassifier.scala

+
+    val pixels = new ListBuffer[Float]()
+
+    for (x <- 0 until h) {


loops should be based on layout of the input descriptor. find out if H is before W

nswamy · 2018-03-15T18:54:28Z

scala-package/infer/src/main/scala/ml/dmlc/mxnet/infer/ImageClassifier.scala

+      }
+    }
+
+    val reshaped_pixels = NDArray.array(pixels.toArray, shape = Shape(224, 224, 3))


Shape based on input descriptor

nswamy · 2018-03-15T18:55:15Z

scala-package/infer/src/main/scala/ml/dmlc/mxnet/infer/ImageClassifier.scala

+
+    val reshaped_pixels = NDArray.array(pixels.toArray, shape = Shape(224, 224, 3))
+
+    val swapped_axis = NDArray.swapaxes(reshaped_pixels, 0, 2)


would you need to swap if you build the Input according to the input descriptor?

nswamy · 2018-03-15T18:56:35Z

scala-package/infer/src/main/scala/ml/dmlc/mxnet/infer/ImageClassifier.scala

+                        topK: Option[Int] = None): IndexedSeq[List[(String, Float)]] = {
+
+    val width = inputDescriptors(0).shape(2)
+    val height = inputDescriptors(0).shape(3)


same as above.

nswamy · 2018-03-15T18:58:04Z

scala-package/infer/src/main/scala/ml/dmlc/mxnet/infer/ImageClassifier.scala

+    * @param resizedImage BufferedImage to get pixels from
+    * @return NDArray pixels array
+    */
+  def getPixelsFromImage(resizedImage: BufferedImage): NDArray = {


take layout parameter as input and make it static

nswamy · 2018-03-15T18:58:53Z

scala-package/infer/src/main/scala/ml/dmlc/mxnet/infer/ImageClassifier.scala

+
+    val output = super.classifyWithNDArray(input, topK)
+
+    IndexedSeq(output(0))


dispose input NDArray before return the result.

nswamy · 2018-03-15T19:02:25Z

@marcoabreu currently we don't have any hooks to run Scala integration tests. We will take up adding examples to CI later.

marcoabreu · 2018-03-16T12:58:12Z

@nswamy I'm afraid I don't understand. We already have a Scala-Job in CI. Could you elaborate?

nswamy · 2018-03-16T13:22:54Z

@marcoabreu there is no CI job for running integration tests for Scala examples or tests and I am not comfortable adding this to the unit test stage that is why I said this should be taken up separately and don't spend cycles here. Also I will need your help to set up a CI job for running Scala integration tests.

marcoabreu · 2018-03-16T14:57:04Z

I see. Do you think we can add them before the release is cut? I'd be happy to assist in creating the integration job. For the time being (until we got proper nightly tests), we can make a job in the PR pipeline and move it later on.

nswamy · 2018-03-16T18:57:40Z

@marcoabreu I don't think we'll have the time to make this improvement before the release, but surely have it on our plate to take it up soon after.

marcoabreu · 2018-03-16T20:44:34Z

But we can't really announce a new feature without having test coverage :/

nswamy · 2018-03-16T20:47:01Z

@marcoabreu this is an example, are we testing python examples. there are already tests in this PR

nswamy · 2018-03-16T20:50:38Z

@marcoabreu, can you help create a CI Integration test job for Scala ?. We can add it there. I guess its a good thing to make sure the examples work.

marcoabreu · 2018-03-16T21:33:44Z

Sure thing! Would you want me to create a PR to your branch or just explain you here what you should do?

nswamy · 2018-03-16T21:37:36Z

could you please create a PR, we can pick up.

marcoabreu · 2018-03-16T22:31:46Z

Sure, no problem. Can you make the changes in order to allow unit testing and integration testing of the scala package and I'll make the integration into CI? At the moment, maven verify just runs all tests - I guess you're more familiar with the setup on scala side than I am.

See https://github.com/apache/incubator-mxnet/blob/master/scala-package/pom.xml#L254 for reference

Roshrini · 2018-03-16T22:51:07Z

@marcoabreu I will look into it. WIll try to run the example as a part of integration test

aaronmarkham

A PR for the README.md is forthcoming...

aaronmarkham · 2018-03-21T18:06:37Z

scala-package/examples/scripts/inferexample/run_predictor_example.sh

+CLASS_PATH=$MXNET_ROOT/scala-package/assembly/osx-x86_64-cpu/target/*:$MXNET_ROOT/scala-package/examples/target/*:$MXNET_ROOT/scala-package/examples/target/classes/lib/*:$MXNET_ROOT/scala-package/infer/target/*
+
+# model dir
+MODEL_DIR=$1


Can you have this print out usage info?

aaronmarkham · 2018-03-21T18:08:20Z

...kage/examples/src/main/scala/ml/dmlc/mxnetexamples/inferexample/ImageClassifierExample.scala

+  def runInference(modelPathPrefix: String, inputImagePath: String, inputImageDir: String):
+  IndexedSeq[IndexedSeq[(String, Float)]] = {
+    val dType = DType.Float32
+    val inputShape = Shape(1, 3, 224, 224)


In the SSD example we're getting artifacts by a sh script and these include the signature.json. Shouldn't we be fetching the input info from this?

aaronmarkham · 2018-03-21T18:09:11Z

...kage/examples/src/main/scala/ml/dmlc/mxnetexamples/inferexample/ImageClassifierExample.scala

+  @Option(name = "--model-dir", usage = "the input model directory")
+  private val modelPathPrefix: String = "/resnet/resnet-152"
+  @Option(name = "--input-image", usage = "the input image")
+  private val inputImagePath: String = "/images/Cat-hd-wallpapers.jpg"


prefer to keep the test image artifact consistent: kitten.jpg (plus this is easily grabbed from s3 with some links we have in the MMS repo).

aaronmarkham · 2018-03-21T18:33:47Z

...examples/src/test/scala/ml/dmlc/mxnetexamples/inferexample/ImageClassifierExampleSuite.scala

+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+


Usage info?

aaronmarkham · 2018-03-21T18:35:33Z

...examples/src/test/scala/ml/dmlc/mxnetexamples/inferexample/ImageClassifierExampleSuite.scala

+    "wget http://data.mxnet.io/models/imagenet/resnet/synset.txt -P resnet/ -q --show-progress"!
+
+    "wget " +
+      "http://thenotoriouspug.com/wp-content/uploads/2015/01/Pug-Cookie-1920x1080-1024x576.jpg " +


can we use something on s3 in case this image just disappears for some reason?

aaronmarkham · 2018-03-21T18:36:30Z

scala-package/infer/src/test/scala/ml/dmlc/mxnet/infer/ImageClassifierSuite.scala

+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+


usage info?

nswamy · 2018-03-21T19:18:45Z

scala-package/examples/pom.xml

-    <version>1.2.0-SNAPSHOT</version>
-    <relativePath>../pom.xml</relativePath>
-  </parent>
+    <modelVersion>4.0.0</modelVersion>


remove tabs and replace with spaces

nswamy · 2018-03-21T19:25:45Z

scala-package/infer/src/main/scala/ml/dmlc/mxnet/infer/ImageClassifier.scala

+
+  val inputShape = inputDescriptors(0).shape
+
+  // Considering 'NCHW' as default layout when not provided


you need force user to pass the shape and layout for CHW, you can only use Batch size defaulted to the first axis when not passed

nswamy · 2018-03-21T19:26:17Z

scala-package/infer/src/main/scala/ml/dmlc/mxnet/infer/ImageClassifier.scala

+  def classifyImage(inputImage: BufferedImage,
+                    topK: Option[Int] = None): IndexedSeq[IndexedSeq[(String, Float)]] = {
+
+    val scaledImage = ImageClassifier.reshapeImage(inputImage, width, height)


also transpose if the layout is different, example channel is last. Lets do that as a separate task

nswamy · 2018-03-21T19:27:45Z

scala-package/infer/src/main/scala/ml/dmlc/mxnet/infer/ImageClassifier.scala

+    val op = NDArray.concatenate(imageBatch)
+
+    val result = super.classifyWithNDArray(IndexedSeq(op), topK)
+


also dispose the list of NDArrays imageBatch

nswamy · 2018-03-21T19:33:48Z

scala-package/infer/src/main/scala/ml/dmlc/mxnet/infer/ImageClassifier.scala

+  }
+
+  /**
+    * Read image file from provided path


nswamy · 2018-03-21T19:36:44Z

scala-package/infer/src/main/scala/ml/dmlc/mxnet/infer/ImageClassifier.scala

+      var col = 0
+      while (col < w) {
+        val rgb = pixels(row * w + col)
+        result(0 * h * w + row * w + col) = (rgb >> 16) & 0xFF


can you add comments

nswamy · 2018-03-21T19:37:37Z

scala-package/infer/src/main/scala/ml/dmlc/mxnet/infer/ImageClassifier.scala

+    *
+    * @param resizedImage BufferedImage to get pixels from
+    * @param inputImageShape Should be same as inputDescriptor shape
+    * @return NDArray pixels array


Add a Note saying Caller is responsible to dispose the returned NDArray

nswamy · 2018-03-21T19:38:17Z

scala-package/infer/src/main/scala/ml/dmlc/mxnet/infer/ImageClassifier.scala

+    val imageBatch = ListBuffer[NDArray]()
+    for (image <- inputBatch) {
+      val scaledImage = ImageClassifier.reshapeImage(image, width, height)
+      val pixelsNdarray = ImageClassifier.bufferedImageToPixels(scaledImage, inputShape)


dispose pixelsNDArray

nswamy

very good job Roshani!
last few changes before we can merge this.

nswamy · 2018-03-21T21:53:02Z

scala-package/infer/src/main/scala/ml/dmlc/mxnet/infer/ImageClassifier.scala

@@ -44,12 +45,19 @@ class ImageClassifier(modelPathPrefix: String,

  val classifier: Classifier = getClassifier(modelPathPrefix, inputDescriptors)

+  require(inputDescriptors.head.shape.length != 0,


.length >= 3

this check already comes from DataDesc. So, wasnt adding it here. But yeah I can add extra check

nswamy · 2018-03-21T21:53:43Z

scala-package/infer/src/main/scala/ml/dmlc/mxnet/infer/ImageClassifier.scala

+  require(inputDescriptors.head.shape.length != 0,
+    "Please provide shape information in the descriptor")
+
+  require(!inputDescriptors.head.layout.isEmpty,


no need for the I already changed the datadescriptor to have shape.length == layout.length

nswamy · 2018-03-21T21:54:34Z

scala-package/infer/src/main/scala/ml/dmlc/mxnet/infer/ImageClassifier.scala

  val inputLayout = inputDescriptors(0).layout

  val inputShape = inputDescriptors(0).shape

  // Considering 'NCHW' as default layout when not provided
  // Else get axis according to the layout
+  // [TODO] if layout is different


if layout is different than the bufferedImage layout, transpose to match the inputdescriptor shape

nswamy · 2018-03-21T21:55:55Z

scala-package/infer/src/main/scala/ml/dmlc/mxnet/infer/ImageClassifier.scala

@@ -122,31 +133,45 @@ object ImageClassifier {
  }

  /**
-    * Read image file from provided path
+    * Convert input BufferedImage to NDArray of input shape
+    *


start the note with <p> and end with <p>

nswamy · 2018-03-21T21:56:54Z

scala-package/examples/pom.xml

@@ -147,5 +174,12 @@
      <artifactId>opencv</artifactId>
      <version>2.4.9-7</version>
    </dependency>
+    <!-- https://mvnrepository.com/artifact/org.mockito/mockito-all -->


do you need mockito for testing the example. this should be like a integration test

nswamy · 2018-03-21T22:10:38Z

scala-package/infer/src/main/scala/ml/dmlc/mxnet/infer/ImageClassifier.scala

+
+  require(!inputDescriptors.head.layout.isEmpty,
+    "Please provide layout information in the descriptor")
+
  val inputLayout = inputDescriptors(0).layout


can you make the below variables protected[infer]

aaronmarkham · 2018-03-22T20:03:07Z

Docs GTG.

marcoabreu

Added my review. @aaronmarkham @nswamy are your comments still valid?

marcoabreu · 2018-03-22T22:11:31Z

scala-package/examples/pom.xml

@@ -14,6 +14,24 @@
  <name>MXNet Scala Package - Examples</name>

  <profiles>
+    <profile>
+      <id>osx-x86_64-cpu</id>


Is the default behaviour windows or why is it not defined here?

there is no support for windows in the current Scala-package and there are no plans to add with this change. This has to be taken up as a separate task for MXNet-Scala binding

Ah I wasn't aware of that, thanks for elaborating

marcoabreu · 2018-03-22T22:13:07Z

scala-package/examples/scripts/inferexample/imageclassifier/run_classifier_example.sh

+# KIND, either express or implied.  See the License for the
+# specific language governing permissions and limitations
+# under the License.
+


Please add set -e (to allow proper failure detection in automated environments)

marcoabreu · 2018-03-22T22:14:27Z

...t/scala/ml/dmlc/mxnetexamples/inferexample/imageclassifier/ImageClassifierExampleSuite.scala

+      "/resnet18/ -q") !
+
+    Process("wget " +
+      "http://thenotoriouspug.com/wp-content/uploads/2015/01/Pug-Cookie-1920x1080-1024x576.jpg " +


Could this be moved to S3?

marcoabreu · 2018-03-22T22:15:28Z

scala-package/infer/pom.xml


-    <artifactId>mxnet-infer</artifactId>
+    <artifactId>mxnet-infer_2.11</artifactId>


What exactly is the "2.11" here? To me, it sounds like this is some kind of centralized version. Maybe use a variable instead? (also above). Is this related to "{scala.binary.version}"?

It is the Scala binary version, this is how other modules in this package are built. We won't be refactoring the entire project here.

marcoabreu · 2018-03-22T22:19:07Z

scala-package/infer/src/test/scala/ml/dmlc/mxnet/infer/ImageClassifierSuite.scala

+    val image1 = new BufferedImage(100, 200, BufferedImage.TYPE_BYTE_GRAY)
+    val image2 = ImageClassifier.reshapeImage(image1, 1000, 2000)
+
+    assert(image2.getWidth === 1000)


Could you execute this test once, generate an MD5 hash based on that output and verify it here? The shape will probably always be right, but the question is whether the actual content has been resized correctly or the code just generated some random data.

Python has libraries to do image hashing(imagehash). Getting hash is not that trivial in Java. Will have to implement image hashing function (average_hash)to match the hash.

However, I ran the function "ImageClassifier.reshapeImage" and tested the hash with python if it's resizing correctly or not. It's working as expected.

@marcoabreu I am not convinced that it is worth the effort and extra code to test this functionality. see this https://www.pyimagesearch.com/2017/11/27/image-hashing-opencv-python/.
Also, the reshaped images are used to test against the model for inference and we verify the output(label) matches what the expected output.

Oh I see, validating them in a follow up step with the output labels is even better

marcoabreu · 2018-03-22T22:20:31Z

scala-package/infer/src/test/scala/ml/dmlc/mxnet/infer/ImageClassifierSuite.scala

+
+    val result = ImageClassifier.bufferedImageToPixels(image2, Shape(1, 3, 2, 2))
+
+    assert(result.shape == inputDescriptor(0).shape)


Could you verify the content? (See above)

marcoabreu · 2018-03-22T22:22:23Z

scala-package/examples/pom.xml

@@ -14,6 +14,24 @@
  <name>MXNet Scala Package - Examples</name>

  <profiles>
+    <profile>


What happens if an undefined platform is being detected?

Only the mentioned platforms are supported similar to what the current scala-package supports.

marcoabreu

Lgtm, Thanks for addressing my comments so quickly. Good job!

* Image classifier for infer package

Roshrini requested a review from yzhliu as a code owner March 9, 2018 23:44

Roshrini changed the title ~~[WIP] Image classifier for scala-infer package~~ [MXNET-50][WIP] Image classifier for scala-infer package Mar 9, 2018

nswamy changed the title ~~[MXNET-50][WIP] Image classifier for scala-infer package~~ [MXNET-53][WIP] Image classifier for scala-infer package Mar 10, 2018

nswamy requested review from nswamy and CodingCat March 10, 2018 00:14

nswamy reviewed Mar 10, 2018

View reviewed changes

marcoabreu suggested changes Mar 10, 2018

View reviewed changes

nswamy suggested changes Mar 15, 2018

View reviewed changes

Roshrini force-pushed the scala-infer-package branch 3 times, most recently from b38551a to fd2c2ae Compare March 21, 2018 16:52

aaronmarkham suggested changes Mar 21, 2018

View reviewed changes

Roshrini force-pushed the scala-infer-package branch from c36b6a8 to 22fcac1 Compare March 21, 2018 19:22

nswamy suggested changes Mar 21, 2018

View reviewed changes

Roshrini force-pushed the scala-infer-package branch from ba3a43f to 388fb4d Compare March 21, 2018 21:56

nswamy suggested changes Mar 21, 2018

View reviewed changes

nswamy reviewed Mar 21, 2018

View reviewed changes

Roshrini force-pushed the scala-infer-package branch from abf4688 to f7e70c4 Compare March 22, 2018 00:12

Nagmote and others added 15 commits March 22, 2018 12:37

rearrangement

caa1450

remove duplicate files

dbe013f

comments added

b605594

review comments addressed

197e6d2

indentation fixed

e99b962

version added to infer package pom

79d620c

first round of edits pt 2

1bcde1d

few more comments addressed

7f40f24

changing temp directory

3add55a

input image changed to one from s3

923cea1

slf4j error resolved

82763f2

temp dir fix

d305933

fixes

77198ec

renaming prefix

ed3a527

model path updated

6b3c6ec

Roshrini force-pushed the scala-infer-package branch from 3d5659c to 6b3c6ec Compare March 22, 2018 19:37

aaronmarkham added 2 commits March 22, 2018 13:00

tiny table update

02fbc8b

it is all relative

ec9bb45

aaronmarkham mentioned this pull request Mar 22, 2018

[MXNET-116] object detector class added #10179

Closed

6 tasks

marcoabreu suggested changes Mar 22, 2018

View reviewed changes

some fixes

b440f4f

marcoabreu approved these changes Mar 23, 2018

View reviewed changes

formatting

95ca571

nswamy merged commit f386d91 into apache:master Mar 23, 2018

ashokei pushed a commit to ashokei/incubator-mxnet that referenced this pull request Mar 27, 2018

[MXNET-53]Image classifier for scala-infer package (apache#10054)

2d8fa04

* Image classifier for infer package

jinhuang415 pushed a commit to jinhuang415/incubator-mxnet that referenced this pull request Mar 30, 2018

[MXNET-53]Image classifier for scala-infer package (apache#10054)

45a6bbe

* Image classifier for infer package

rahul003 pushed a commit to rahul003/mxnet that referenced this pull request Jun 4, 2018

[MXNET-53]Image classifier for scala-infer package (apache#10054)

0a98392

* Image classifier for infer package

zheng-da pushed a commit to zheng-da/incubator-mxnet that referenced this pull request Jun 28, 2018

[MXNET-53]Image classifier for scala-infer package (apache#10054)

9247538

* Image classifier for infer package

Roshrini deleted the scala-infer-package branch November 15, 2018 18:18


		val reshaped_pixels = NDArray.array(pixels.toArray, shape = Shape(224, 224, 3))

		val swapped_axis = NDArray.swapaxes(reshaped_pixels, 0, 2)


		val output = super.classifyWithNDArray(input, topK)

		IndexedSeq(output(0))


		val inputShape = inputDescriptors(0).shape

		// Considering 'NCHW' as default layout when not provided

		val op = NDArray.concatenate(imageBatch)

		val result = super.classifyWithNDArray(IndexedSeq(op), topK)

		@@ -44,12 +45,19 @@ class ImageClassifier(modelPathPrefix: String,

		val classifier: Classifier = getClassifier(modelPathPrefix, inputDescriptors)

		require(inputDescriptors.head.shape.length != 0,


		<artifactId>mxnet-infer</artifactId>
		<artifactId>mxnet-infer_2.11</artifactId>


		val result = ImageClassifier.bufferedImageToPixels(image2, Shape(1, 3, 2, 2))

		assert(result.shape == inputDescriptor(0).shape)

[MXNET-53]Image classifier for scala-infer package #10054

[MXNET-53]Image classifier for scala-infer package #10054

Conversation

Roshrini commented Mar 9, 2018 • edited Loading

Description

Checklist

Essentials

Changes

Comments

Choose a reason for hiding this comment

marcoabreu left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

nswamy commented Mar 15, 2018

marcoabreu commented Mar 16, 2018

nswamy commented Mar 16, 2018

marcoabreu commented Mar 16, 2018

nswamy commented Mar 16, 2018

marcoabreu commented Mar 16, 2018

nswamy commented Mar 16, 2018

nswamy commented Mar 16, 2018 • edited Loading

marcoabreu commented Mar 16, 2018

nswamy commented Mar 16, 2018

marcoabreu commented Mar 16, 2018

Roshrini commented Mar 16, 2018

aaronmarkham left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

nswamy left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

aaronmarkham commented Mar 22, 2018

marcoabreu left a comment • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

marcoabreu left a comment

Choose a reason for hiding this comment

Roshrini commented Mar 9, 2018 •

edited

Loading

nswamy commented Mar 16, 2018 •

edited

Loading

marcoabreu left a comment •

edited

Loading