• Title: Marine mammal voice recognition - right-whale-redux
• Author: Yiquan Lin (linyq78@mail2.sysu.edu.cn)
• Submission date: 7th Jan, 2019
• Full text: Digital version

• Dataset:

The ICML 2013 Whale Challenge - Right Whale Redux

• Dataset structure after unzip:

 + all
   + test2
   + train2
   + sampleSubmission.csv

• Python environment (Python 3.6.7):

tensorflow==1.12.0
scikit-learn==0.20.2
scikit-image==0.14.1
opencv-python==3.4.5.20
scipy==1.2.0

• For installing above modules, bash in ./ and run:

pip install -r requirements.txt

• Ensure you correctlty install all modules, run input_processor_vis.py in shell:

python input_processor_vis.py

• following operation has been encapsulated in _read_and_preprocess function in input_processor.py. In training phase, func will called by tf.data and tf.estimator automatically.

• Unzip downloaded dataset to ./data, then run annota_generator.py in shell for splitting train/val and generating annotation:

python annota_generator.py

the train/val annotation will be saved at ./data.

• Finally we get follow directory structure:

  + data
    + model
    + test2
    + train2
    + train_annotation.txt
    + eval_annotation.txt
  + annota_generator.py
  + input_processor.py
  + input_processor_vis.py
  + train.py
  + requirements.txt
  

• Run train.py in shell, and model checkpoint/events/results will be saved ./data/model (when training model on the splited train set and evaluate model on validation set).

　　[1] Peter J. Dugan, Christopher W. Clark, Yann André LeCun, Sofie M. Van Parijs- DCL System Using Deep Learning Approaches for Land-based or Ship-based Real-Time Recognition and Localization of Marine Mammals

　　[2] Mohammad Pourhomayoun, Peter Dugan, Marian Popescu, Christopher Clark - Bioacoustic Signal Classification Based on Continuous Region Processing, Grid Masking and Artificial Neural Network