ktl014 / mpl

Dataset augmentation for improving biological specimen classification. Coded in Python and used Caffe framework (Scripps Institution Oceanography Internship Summer 17)

Home Page:https://ktl014.github.io/mpl/

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Marine Physical Laboratory Summer 2017 Internship

Welcome! This is my summer internship work page during my time at the Marine Physical Laboratory at the Scripps Insitution of Oceanography under Doctor Jules Jaffe and his PhD students. My research is primarily focused on applying computer vision and machine learning techniques to efficiently analyze the big data that is provided through their in situ optical underwater imaging systems.

Click here to view my latest work

Overview

We have chosen to experiment with three source domains to test our hypothesis on utilizing a different source domain to boost classification performance and reduce the amount of effort to train the classifiers. Below are our source domains:

  1. SPCBench
  2. SPCInsitu
  3. SPCombo

We will evaluate these three source domains' experiments by testing them on our target domain dataset, which will ultimately provide us an accurate representation of how these classifiers fare in a realistic deployment environment.

Project Presentation & Report

Below is a presentation & report of all my experiments and their results to better explain my research.

Mpl summer 2017 presentation

Mpl summer 2017 project report

Python scripts

The eval.py and train_alexnet.py script are definitive python scripts that are robust enough to use for all of the source domain experiments, as well as evaluating the target domain. More information regarding how to navigate through these scripts will come!

For More Information:

Keep a look out for the conference release of this research at the 2018 Ocean Sciences Meeting from February 11-16th

Feel free to contact me if you'd like to hear about my research. My contact information, as well as my resume and LinkedIn, can be found below

Kevin Le

kevin.le@gmail.com | LinkedIn | Resume

About

Dataset augmentation for improving biological specimen classification. Coded in Python and used Caffe framework (Scripps Institution Oceanography Internship Summer 17)

https://ktl014.github.io/mpl/


Languages

Language:Jupyter Notebook 98.1%Language:Python 1.9%