Script to classify images of plants and animals with the image-based species recognition models created in the context of the paper by Van Horn et al. 2021, benchmarking different kinds of models created on basis of the dataset of the iNaturalist 2021 Comptetition. The code was extracted from this repository by deblagoj as a contribution to the iNaturalist 2017 Competition.
-
(recommended but not absolutely necessary) create and activate own (conda) environment
-
install packages
pip install torch torchvision
2.1 Actually this should work, otherwise you can also use the provided conda environment file environment.yml.
Run in the repository:
conda env create -f environment.yml
- Clone this repository
git clone https://github.com/EibSReM/iNaturalist_Competition.git
and change to respective directory
-
Download pretrained models from the paper here (5,6 GB), mentioned in the papers repository.
-
Unzip file with pretrained models (probably needs to be unzipped twice: first time the .tar.gz file, sedondly the .tar file in the unzipped folder from the previous step)
-
Adapt path to pytorch model in the
inference.pyscript (we used the model in: cvpr21_newt_pretrained_models\cvpr21_newt_pretrained_models\pt\inat2021_supervised_large_from_scratch.pth.tar) -
Adapt path to images (folder) in the
inference.pyscript -
Run script
python inference.py
- Find results in
Output.txt
Classifying 258 images, the script run 81.02 seconds on a Windows 10 notebook with the following hardware specifications:
- Intel(R) Core(TM) i7-8565U CPU @ 1.80GHz 1.99 GHz
- 20 GB Ram
In the folder demo you can find a couple of images and a README with the expected results for demonstration purposes.
Van Horn G, Cole E, Beery S, Wilber K, Belongie S, Mac Aodha O, et al. (2021) Benchmarking Representation Learning for Natural World Image Collections. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition 12884‑12893. http://arxiv-export-lb.library.cornell.edu/pdf/2103.16483