Differences

This shows you the differences between two versions of the page.

--- cs501r_f2016:tmp [2016/09/19 17:02]
wingated
+++ cs501r_f2016:tmp [2016/11/09 18:21]
wingated
@@ Line 1: / Line 1: @@
 ====Objective:====
-To explore deeper networks, to leverage convolutions, and to explore Tensorboard.
+To gain experience coding a DNN architecture and learning program end-to-end, and to gain experience with Siamese network and ResNets.
 ----
 ====Deliverable:====
-For this lab, you will need to perform three steps:
+For this lab, you will need to implement a simple face similarity detector.
-  - You need to implement the [[https://www.tensorflow.org/versions/r0.10/tutorials/index.html|Deep MNIST for experts tutorial]]
+  - You must implement a siamese network that accepts two input images
-  - You need to modify the tutorial code to deliver visualizations via Tensorboard.
+  - The network must output the probability that the two images are the same class
+  - Your implementation should use a ResNet architecture
-Specifically, you should  turn in an iPython notebook that shows two images:
+You should turn in the following:
-  - A Tensorboard image showing your classification accuracy over time
-  - A Tensorboard image showing your (expanded) computation graph
-An example of the classification accuracies are shown at the right.  An example of the expanded computation graph is shown down below.  (Note the "Download PNG" button in the upper-left of Tensorboard!)
+  - A tensorboard screenshot showing that your architecture is, indeed, a siamese architecture
+  - Your code
+  - A small writeup (<1/2 page) describing your test/training split, your resnet architecture, and the final performance of your classifier.
-According to the tutorial, if you run for 20,000 iterations, the final accuracy of your classifier will be around 99.5%.  To make your life simpler, you only need to run for 1500 iterations.
+You should use the [[http://www.openu.ac.il/home/hassner/data/lfwa/|Labeled Faces in the Wild-a]] dataset (also available for
+[[http://liftothers.org/byu/lfwa.tar.gz|download from liftothers]]).
 ----
@@ Line 24: / Line 26: @@
 Your notebook will be graded on the following:
-  * 40% Correct multilayer convolutional network defined and working
+  * 35% Correct implementation of Siamese network
-  * 30% Tidy and legible display of Tensorboard accuracy
+  * 35% Correct implementation of Resnet
-  * 30% Tidy and legible display of Tensorboard computation graph
+  * 20% Reasonable effort to find a good-performing topology
+  * 10% Results writeup
 ----
 ====Description:====
-You now understand the basics of multi-layer neural networks.  Here, we'll expand on your toolkit by adding in convolutions, a bit of dropout, and a new optimization method.  Most of these will be explained in future lectures, so for now we will just use them without (fully) understanding them.
+----
+====Hints:====
-**Part 1: implement deep convolutional networks **
+To help you get started, here's a simple script that will load all of the images and calculate labels.  It assumes that the face database has been unpacked in the current directory, and that there exists a file called ''list.txt'' that was generated with the following command:
-For this lab, you must implement the [[https://www.tensorflow.org/versions/r0.10/tutorials/index.html|Deep MNIST for experts tutorial]].  This is mostly cutting-and-pasting code; since you already have Tensorflow up and running, this should be fairly straightfoward.
+<code bash>
+find ./lfw2/ -name \*.jpg > list.txt
+</code>
-A few things to note:
+After running this code, the data will in the ''data'' tensor, and the labels will be in the ''labels'' tensor:
-  - You are now adding multiple layers.  Be careful with your variable names!
+<code python>
-  - You'll use the Adam optimizer, not vanilla SGD.  We learn more about this later.
-  - The dropout layer is optional, but you should probably leave it in just to make cutting-and-pasting easier.
-**Note:** you only need to train for 1500 steps.  My final accuracy was 96.5%.  If you want to train for the full 20k steps, you are of course welcome to do so!
+from PIL import Image
+import numpy as np
-**Part 2: add in Tensorboard visualizations**
+#
+# assumes list.txt is a list of filenames, formatted as
+#
+# ./lfw2//Aaron_Eckhart/Aaron_Eckhart_0001.jpg
+# ./lfw2//Aaron_Guiel/Aaron_Guiel_0001.jpg
+# ...
+#
-{{ :cs501r_f2016:graph-run_2_.png?direct&200|}}
+files = open( './list.txt' ).readlines()
-There are two parts to this: first, you need to scope all of the nodes in your computation graph.  In class, I showed a visualization that drew pretty boxes around all of the different parts of your computation graph.  That's what I want from you!  Check out the ''tf.name_scope'' function.  You should create 4 scopes: "Wx_B" for your computation graph, "Cost" for your cost function, "accuracy" for your accuracy calculations, and "optimizer" for your optimizer.  When you're done, you should have something that looks like the graph on the introduction.
+data = np.zeros(( len(files), 250, 250 ))
+labels = np.zeros(( len(files), 1 ))
-Second, you'll need to produce little graphs that show accuracy over time.  This is done by asking Tensorflow to create summaries of key variables.  Adventurous souls can dive right into the [[https://www.tensorflow.org/versions/r0.10/how_tos/summaries_and_tensorboard/index.html|Tensorflow visualization tutorial]].  Here are some condensed notes:
+# a little hash map mapping subjects to IDs
+ids = {}
+scnt = 0
-Tensorboard logs //events// to a //summary log//.  You'll need to tell Tensorboard where to stash those events and when to write them out; both are done with a SummaryWriter.  You need to create a SummaryWriter object:
+# load in all of our images
+ind = 0
+for fn in files:
-''summary_writer = tf.train.SummaryWriter( "./tf_logs", graph=sess.graph )''
+    subject = fn.split('/')[3]
+    if not ids.has_key( subject ):
+        ids[ subject ] = scnt
+        scnt += 1
+    label = ids[ subject ]
+    data[ ind, :, : ] = np.array( Image.open( fn.rstrip() ) )
+    labels[ ind ] = label
+    ind += 1
-as well as scalar summaries of relevant variables; maybe something like this:
+# data is (13233, 250, 250)
+# labels is (13233, 1)
-''acc_summary = tf.scalar_summary( 'accuracy', accuracy )''
-These summaries are considered ops, just like any node in the computation graph, and they are triggered by ''sess.run''.  Tensorflow helpfully allows you to merge all of the summary ops into a single operation:
-''merged_summary_op = tf.merge_all_summaries()''
-Then, you'll need to trigger the ''merged_summary_op'' operation.  This will generate a //summary string//, which you should pass to your summary writer.
-Once you have run your code and collected the necessary statistics, you should be able to start up the Tensorboard visualizer.  It runs as a webserver; to start Tensorboard, you should be able to run something like the following **from the directory where you ran your TF code**:
-<code bash>
-cd tf_logs
-tensorboard --logdir .
 </code>
-At which point you'll see something like the following output:
-<code>
-Starting TensorBoard 28 on port 6006
-(You can navigate to http://192.168.250.107:6006)
-</code>
-Point your browser to the spot indicated, and voila!
-----
-====Hints:====

BYU CS classes

User Tools

Site Tools

Differences

Page Tools