This is the project repository for a TensorFlow workshop in Chicago in May, 2018.

• Schedule
• Sessions
• Slack
• Servers

• 10:00 - 10:30 (30 min) : Registration + setup
• 10:30 - 11:10 (40 min) : Session 1
• 11:10 - 11:25 (15 min) : Break
• 11:25 - 12:05 (40 min) : Session 2
• 12:05 - 12:20 (15 min) : Break
• 12:20 - 13:00 (40 min) : Session 3
• 13:00 - 14:00 (60 min) : Lunch
• 14:00 - 14:40 (40 min) : Session 4
• 14:40 - 14:55 (15 min) : Break
• 14:55 - 15:35 (40 min) : Session 5
• 15:35 - 15:50 (15 min) : Break
• 15:50 - 16:30 (40 min) : Session 6
• 16:30 - 17:00 (30 min) : Wrap up + spill over

Below is a working list of workshop session topics.

Objective: Walk students through the process we'll be going through during the workshop

This will include an introduction to Guild and the workshop project. We can pair off and access an EC2 instance to run various operations:

• Acquire images
• Detect classes
• Run model as application

Objective: Understanding the main concepts around image classification (convolutional neural networks, transfer learning)

Will share a simple notebook around cats/dogs just for learning purposes

Objective: A quick tour of the main concepts developed in the last few years in object detection, ending with Mask-RCNN in Keras

Architectures include: R-CNN Fast R-CNN Faster R-CNN Mask R-CNN One shot (YOLO, SSD)

Will share other repos of other approaches and walk through code examples of the main concepts.

Objective: Fine-tune a pretrained model on a narrower set of images

E.g. use pets database to fine-tune a model trained on COCO.

Students should see some difference between the application demonstrated in Session 2 - ideally it would be an improvement.

Objective: Generate a dataset unique to the room and object being detected

Objective: Fine-tune a pretrained model on the dataset generated in Session 5

Students should see a marked improvement in the detector.

We will use Slack to share information during the workshop and for any followup work.

Students will receive an invitation to Slack before the workshop begins.

Server hostnames and user passwords will be assigned via the Slack channel during the workshop.

Once you have your server hostname, follow these steps to log on.

• In your browser, open https://<server hostname>
• Log on with user name ubuntu
• Specify the password provided

NOTE: You must use https in the browser URL above. If you use http the browser will not connect to the server.

This will open a terminal session on the server. You can repeat these steps in a new tab or window to open multiple terminal sessions to a server.

If you are logging in for the first time, start a tmux session by running:

tmux

tmux is a terminal multiplexer that lets you open multiple terminals in a single session. It also lets you reconnect to your terminals if you get disconnected.

If you already have a tmux session, attach to it by running:

tmux attach

Each time you open a terminal session to a server, including a new tmux window, you must activate the environment used for training and other operation.

To activate the workshop environment, in a terminal session, run:

source activate workshop

All tmux commands are run by typing Ctrl-b followed by a command character.

Ctrl-b c

Ctrl-b w

Ctrl-b NUM

Previous:

Ctrl-b p

Next:

Ctrl-b n

Ctrl-b l

Start scrolling:

Ctrl-b PGUP

Exit scrolling:

q
Ctrl-C

Ctrl-b d

guild runs