Part 2 - Inception Graph with Retrained Inception Model (Done on 27th May, 2017)

Next, we wanted to train the model with our own images.

First we tried Tensorflow with this tutorial: tutorial But it only changed the final layer and it was used to classify images. For now we haven’t found out a way to train our own model using Tensorflow. So we tried Caffe in the next step.

For Caffe, here is an amazing tutorial: tutorial

First, to use Caffe, we need an Ubuntu system, version 16.04 is highly recommended. Caffe+Cuda8.0+ Ubuntu16.04 installation tutorial: tutorial

Then follow the tutorial above. Here is an example of the folder ‘MYNET’: example

We trained 1662 images of buildings designed by Frank Gehry with 65k iterations. And used this script to run it: script

Result here:

alt text alt text

———————-Input———————————————————Output—————————-