DL_seg_avian_plumage

This is the repo for the analysis code in Deep learning image segmentation reveals patterns of UV reflectance evolution in passerine birds (He et. al. 2022).

This is the code using DeepLab to segment plumage area of bird images and compare the performance to some classic segmentation methods (thresholding, region growing, graph cut and chan-vese). The model is applied on a dataset of passerine birds. The segmentation results are then used to measure the UV reflectance which is used to analyse the evolution of UV in passerine birds.

Prerequisites

• Python 3

• tensorflow = 1.6.0

• numpy >= 1.17.3

• pandas >= 0.23.4

• opencv-python = 4.1.1.26

• scikit-imgae = 0.16.2

• R = 4.1.0

• raster = 3.4-5

• pavo = 2.6.1

• phangorn = 2.5.5

• MCMCglmm = 2.32

Installation

git clone https://github.com/EchanHe/DL_seg_avian_plumage.git

Usage

Segmentation codes are stored in segmentation_code/ folder. 5 images and their annotations are stored in segmentation_code/data folder, the visualise_data.ipynb can be used to visualise the segmentation.

Classic segmentation methods are implemented in classic_segmentation_methods.ipynb

Config files

Adjust the [Directory] section to fit your workspace

file_col is the column name for the file name

cols_override is the column names for segmentation

The other settings and hyperparameters can be set in the config file as well.

Training

python train.py <training_config>.cfg

Predicting

python pred.py <predict_config>.cfg

Analysis codes are stored in analysis_code/ folder including codes for plotting figures in the paper and analyse the UV reflectance of passerine birds.