Imageomics-Species-ReID

Service supporting reidentification with machine learning for various animal species based on Wildbook's Image Analysis (WBIA) by WildMe

Requirements

• Python 3.11
• Python dependencies listed in environment.yaml

Setup on Windows

• Install a Conda-based Python 3 distribution
• Install Windows Subsystem for Linux with wsl --install or visit Install WSL for further instructions
• Activate WSL with wsl and navigate to the desired directory
• Install Mambaforge:

curl -L https://github.com/conda-forge/miniforge/releases/latest/download/Mambaforge-Linux-x86_64.sh -o Mambaforge-Linux-x86_64.sh
bash Mambaforge-Linux-x86_64.sh

• When you are asked the following question, answer with yes:

Do you wish the installer to prepend the install location to PATH ...? [yes|no]

• Clone the repository:

git clone https://github.com/Ziesski/GGR.git
cd GGR

• Install snakemake and other dependencies within an isolated environment (smk_pipeline can be replaced with an alternative name):

conda activate base
mamba env create --name smk_pipeline --file environment.yaml
conda activate smk_pipeline

Required Files

Add the directory to the input images in the root of the repository.

mkdir test_dataset

Insert the ca classifier and viewpoint classifier models into the test_dataset directory and mark it in config.yaml.

Inser the lca verifier probs into the test_dataset directory and mark it in config.yaml.

Add ground truth annotations to test_dataset if you want to filter animal detections.

You can adjust src paths in config.yaml

GGR 
│ # additional files
├── test_dataset
│   ├── your_viewpoint_classifier.pth
│   ├── your_ca_classifier.pth
│   ├── your_verifiers_probs.json
│   └── intermediate/output folders and files
├── images
│ # repository files
└── ...

Alternative Setup

• Clone the repository:

git clone https://github.com/Ziesski/GGR.git

• Full and alternative instructions for installing snakemake can be found in the snakemake documentation (Installation | Setup)

Execution

• Activate WSL with wsl and base conda environment with conda activate base if not active
• Activate snakemake environment (smk_pipeline can be replaced with an alternative name):

conda activate smk_pipeline

• Run the pipeline:

snakemake -s snakefile.smk --cores 1

Final results will appear in test_dataset/output and intermediate results will appear in labeled folders in test_dataset accordingly.

GGR-Specific Functions

• Additional script extrapolate_ggr_gps.py extrapolates from existing GPS data to fill in for missing GPS data for images from other cameras.
• Run in same environment by providing input and output image data .json paths:

python extrapolate_ggr_gps.py test_dataset/image_data.json test_dataset/image_data_complete.json

Tools

YOLO BioCLIP

Future Tasks

• Implement threading for image parameter computation and validity checking if image import is too slow.
• Set up pipeline within IDEA cluster and perform larger tests.