sheppard/CitSci2017_TensorflowRetrainer

CitSci2017 Conference Create Together Day - Project to use Tensorflow to train a model & use it in wq app

CitSci2017_TensorflowRetrainer

CitSci2017 Conference: Create Together Day:

Project to use Tensorflow to train a model & use it in wq app

Basic idea is outlined here:

https://www.tensorflow.org/tutorials/image_retraining

Integration in wq

Here all infos regarding the wq app:

https://wq.io/

https://github.com/wq/wq

Steps to set-up an AWS Ubuntu 16 instance with GPU, docker and tensorflow.

Some of the sources I have used:

https://petewarden.com/2016/09/27/tensorflow-for-mobile-poets/

https://medium.com/@daj/creating-an-image-classifier-on-android-using-tensorflow-part-3-215d61cb5fcd

https://gist.github.com/espdev/a205ec97f650c6695104c3401f7fd9c2

https://alliseesolutions.wordpress.com/2016/09/08/install-gpu-tensorflow-from-sources-w-ubuntu-16-04-and-cuda-8-0/

Set-up a new AWS Ubuntu 16 instance and execute following steps:

Part1_install.sh

Then there is a manual step to get cuDNN:

Part2_install_MANUAL.sh

Finally:

Part3_install.sh

Code to get data from wq server, re-train googles inception model, and upload to wq server

python3 main.py 2 some_description

2: is the id of the dataset from the json db (the dataset must be a zip-file of n folders containing images of n classes)

some description: the description of the final model