/Image-Recognition-Between-Labradoodle-and-Fried-Chicken

Primary LanguagePython

Project: Labradoodle or Fried Chicken? In Black and White.

image

Full Project Description

Term: Spring 2017

  • Team #10

  • Team members

    • Zixuan Guan
    • Senyao Han
    • Yuan Mei
    • Xuanzi Xu
    • Galen Simmons

  • Project summary: In this project, we created a classification engine for grayscale images of poodles versus images of fried chickens. To demonstrate our methods, we created an ipython notebook contained in doc/main_notebook.ipynb. In order to replicate the results we achieved, it is necessary to configure a virtual environment with several key packages including scikit-learn and graphlab-create. Therefore, we provide an environment.yml file, which contains the configuration info necessary to build the virtual environment. Below are instructions.

First, install the virtual environment. Execute: conda env create -f environment.yml

Next, activate the environment. Execute: source activate gl-env

Now, open jupyter notebook: jupyter notebook

In order to process, the images raw images, we provide two python scripts that must be run prior to running the notebook file: resize_images.py and generate_sframe.py. To process a new batch of jpg images, copy the raw jpg images into data/img_raw, and then execute: python resize_images.py python generate_sframe.py Make sure that the above command is executed in the virtual environment. Usually the terminal will display the name of the virtual environment at the prompt in the following format: (gl-env) username$

At this point, there will be images in the data/img_edge, data/img_resize, and data/img_sobel folders. Additionally, there will be SFrame objects saved in the data/sframe folder. We have one last step to run the jupyter notebook. We have to change the base_dir variable in the main_notebook.ipynb file to reflect our local project directory.

Contribution statement: (default) Galen contributed image processing and notebook code and commentary. Xuanzi did neural network linear and kernal SVM and tune the models as well as using principal component analysis to reduce the feature dimensions. Senyao Han, as a presenter, figure out the surf features and hog features for all images, and also wrote the final main.rmd. Zixuan Guan contributed baseline gbm classification, randomForest classification including tuning parameters for these two classification and also xgboost classification. Yuan Mei contributed the tuning of xgboost parameters. All team members contributed equally in all stages of this project. All team members approve our work presented in this GitHub repository including this contributions statement.

Following suggestions by RICH FITZJOHN (@richfitz). This folder is orgarnized as follows.

proj/
├── lib/
├── data/
├── doc/
├── figs/
└── output/

Please see each subfolder for a README file.