WildObs_QLD_WetTropics

The official location of the model files is this Hugging Face repo.

Developers

Prakash Palanivelu Rajmohan and Renuka Sharma

Description

The wildobs_QLD_WetTropics model for Australian Wet Tropics has been trained on ~30k labelled images from 15 classes. We used the Wildlife Observatory of Australia’s tagged image repository for the ‘Wet Tropics’ rainforests (n = 454 camera deployments, 2,184,664 images, 121 species), and refined this to a balanced dataset of the 15 most common species for CV training and testing. We found that fine-tuning SpeciesNet (https://github.com/google/cameratrapai) delivered the highest performance, often exceeding 95% F1-score. We are in the process of publishing our findings as part of this work.

Classes

Alectura lathami or australian brush-turkey <br> Bos taurus or domestic cattle <br> Canis familiaris or domestic dog <br> Casuarius casuarius or southern cassowary <br> Felis catus or domestic cat <br> Heteromyias cinereifrons or grey-headed robin <br> Homo sapiens or human <br> Hypsiprymnodon moschatus or musky rat kangaroo <br> Megapodius reinwardt or orange-footed scrubfowl <br> Orthonyx spaldingii or northern chowchilla <br> Perameles nasuta or long-nosed bandicoot <br> Sus scrofa or wild boar <br> Thylogale stigmatica or red-legged pademelon <br> Uromys caudimaculatus or giant uromys <br> Wallabia bicolor or swamp wallaby <br>

Please build a conda environment using the attached wildobs_QLD_WetTropics_env.yml file before running the evaluation script present in the Python notebook Evaluate_WetTropics_hf.ipynb.

Please unzip the folder images.zip in your directory to be able to use the test files.