https://github.com/blainerothrock/airbus-ship-detection
https://github.com/blainerothrock/airbus-ship-detection
Last synced: 2 months ago
JSON representation
- Host: GitHub
- URL: https://github.com/blainerothrock/airbus-ship-detection
- Owner: blainerothrock
- Created: 2020-03-03T16:26:49.000Z (over 5 years ago)
- Default Branch: master
- Last Pushed: 2020-03-16T21:45:37.000Z (about 5 years ago)
- Last Synced: 2025-02-08T18:30:45.659Z (4 months ago)
- Language: Jupyter Notebook
- Size: 58.9 MB
- Stars: 0
- Watchers: 2
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# Airbus Ship Detection
EE 435: Deep Learning Foundations from Scratch \
Northwestern Univertiy Winter 2020
## Memebers
* Blaine Rothrock
* Ilan Ponsky
* Will Dong
## Overview
Our group is interested in applying knowledge from this course to training TensorFlow models and getting a better understanding of deep artificial neural networks involved in image processing. In order to learn about this process in an organized and efficient way, we utilized a closed Kaggle competition that centered around our topic of image processing with neural networks. The competition we used was the [Airbus Ship Detection Challenge](https://www.kaggle.com/c/airbus-ship-detection). The goal of the Kaggle competition was for participants to be able to build a model that could “detect all ships in satellite images as quickly as possible.”
The specific goals we wanted to hit for our Deep Learning from Scratch final project were to:
* Build a binary classifier model to gain a basic understanding of Tensorflow and how to build models in TensorFlow.
- The objective of the binary classifier model is to output whether an image contained a ship or not utilizing optimization techniques in TensorFlow.
* Explore and implement a U-net model for image segmentation.
- U-Net models are the current state-of-art for image segmentation and where most started for this competition.
- This is a hefty goal given the data size of ~40GB of images and the time it takes to training this complex model. Our goal is to build a model, attempt at training, and gain a understanding of the U-Net architecture.
- To accomplish this, we utilized the notebook of Kevin Mader on Kaggle which served as an excellent foundation to get started with implementing the mentioned goals