Winning the best student paper awar in IW-FCV2018!

Sealion detection and classification

This code was used for NOAA sealion competition which was held in KAGGLE.
Final result is 58th among 385 participants.

Citing this code

Please refer to following paper for detail paper

@inproceedings{IWFCV18sealion,
  title = {Animal Detection in Huge Air-view Images using CNN-based Sliding Window},
  author = {Young-Chul Yoon and Kuk-Jin Yoon},
  booktitle = {International Workshops on Frontiers of Computer Vision},
  year = {2018}
}

Prerequisites

Hardware

Nvidia GPU with at least 3GB memory

Interpreter and Operating System

Python version 2.7 or 3.5
Tested on UBUNTU 16.04 and Windows

Dependency

tensorflow-gpu
python-opencv
imgaug
numpy
csv
skimage

Installing and demo

You have to download this repository and also pre-trained model.

pre-trained model download link : https://drive.google.com/open?id=0Bwaxr_eelTFyS0Vyc2NfajJNb1E

You have to depress the 'sealion_count_model.zip' into 'input' directory.

Running the tests

You can run the control.py (EX. python2 control.py).
The code will show you the detection&classification result on sample images.

About this code

The detector is applied on image through sliding-window method.

The inference is separated into detection and classification stage.

The detailed detection network structure.

Here are several results example of the code.

Acknowledgments

This research is supported by Ministry of Culture, Sports and Tourism(MCST) and Korea Creative Content Agency(KOCCA) in the Culture Technology(CT) Research & Development Program 2017