This repo tests if AI4G models can be deployed using the EcoAssist workflow. In classify.py you can find a minimal reproducable example. Feel free to test by cloning the repo and PR with adjustments.

1. loop through the MegaDetector output JSON to find where the animals are
2. crop out each animal and feed it to the classifier
3. add its predictions to the existing JSON

1. Link to code: At the moment the prediction returns only the highest class. Would it be possible to get all classes with their confidences? Perhaps we can add an extra argument like return_all_predictions to single_image_classification()?

1. It is assumed you have a conda environment called pytorch-wildlife.

   conda create -n pytorch-wildlife python=3.8 -y
   conda activate pytorch-wildlife
   pip install PytorchWildlife
   

2. You have downloaded the model file from Zenodo and placed it in the repo folder: https://zenodo.org/records/10042023/files/AI4GAmazonClassification_v0.0.0.ckpt?download=1
3. Execute script:

   conda activate pytorch-wildlife && python "pytorch-wildlife-test-run\classify.py"