Otolith identification

Install

Install conda and run the following script for local installation

 git clone https://github.com/bacnguyencong/otoliths-identification
 cd otoliths-identification
 conda env create -f environment.yml
 conda activate otoliths-identification-env

Data preprocessing

All references images should be located in ./data/Reference pictures/. Run the following script to split data into training and valid sets:

python preprocess.py

There will be two folders ./data/train and ./data/valid/ containing training and valid data, respectively.

Note: Some images might not be well segmented due to the quality of images, please remove them to clean the data.

Train the model

python main.py -img_size 224 -b 32 -j 4 --arch resnet18 -epochs 100 -lr_patience 5 -early_stop 10 -lr 0.001 --pretrain --train

Predict the labels

python main.py --test

TSNE

python make_tsne.py

Highlight

python visualization.py