bharatkambalur / ece285_s18_project

ECE 285 - Spring 2018 - Machine Learning for Image Processing - Group Project

ECE 285: Image Processing for Machine Learning

Description

This repository contains all the code used for our project "Improving Classification with a Pipelined Architecture using Super-Resolution Methods." The code was developed by Team Perceptive Perceptrons composed of Bharat Kambular, Erik Seetao, Joseph Mattern, and Sharla Chang.

Requirements

The following Python Packages are required: numpy, matplotlib, Pillow, pytorch (tested with 0.3.1)

Install package the packages as follow : $ pip install --user <package_name>

Code organization

• Demos:

  Contains iPython Notebooks that demo different components of our project. Also contains notebooks to produce plots for various test and train accuracies.

• Training:

  Contains the iPython Notebooks that can be used to train DenseNet (Baseline, DBPN Pipelined, Bicubic Scaling Pipelined) and RexNeXt (Baseline only)

• pymodels:

  Contains the Python files that describe the various pytorch models

• models:

  Contains the trained weights for the different networks. (Note: Due to GitHub File size limitation, not all files are uploaded. Please contact the authors to acquire them)

• Results:

  Implementation Specific. Contains the numpy arrays that store various data obtained during training models for each epoch.

• Utils, log:

  Project specific folder that holds Utility scripts and log files respectively