Skip to content

CS 5542 BigData Lab Report #09

Amy Lin edited this page Mar 29, 2017 · 1 revision

TensorFlow PROGRAMMING - CNN

[ QUESTION ]

Implement a CNN model for image classification for the datasets relevant to the project. Report accuracy. Visualizations(Tensor Board): Training and Testing.

[ IMPLEMENTATION ]

Re-scale raw images into same pixel sizes, label and build vectors based on datasets. ( 100 X 100 pixel in this project ) Split data into Test and Training. 70% Training and 30% for Testing. Since we have a small datasets for three of Kansas City Fountains, I set the 10 Testing Images for each group. The rest of the image goes to training. Run 400 iterations to build the CNN model.

< DATASET >

3 Kansas City Fountains

[ RESULTS ]

  • Tensorflow CNN Net




Final Accuracy for 400 Steps : 33%

  • Accuracy







  • Convolution Net 1







  • Convolution Net 2







  • Convolution Net Weight







  • Cross Entropy







  • Dropout ReLU




  • OnConvolution







Visual Question Answering

I haven't been able to work the Heroku and the conversation API for our own KC Fountain Dataset in time! Only put the code up but doesn't have any screenshots yet.

Clone this wiki locally