- Make datasets, look and adapt
make_dataset.sh
, and then run the command. - Visualize the dataset, use
visualize_observation.py
. - Download MAE pretrained model ViT-Large and put it in
checkpoint
directory. - Finetune the model in PlantCLEF2022 dataset, use
finetune.sh
. - Test and predict the results, use
test.sh
, you can see a new directory inresults
directory. - Make the submission result, use
make_official_result.sh
you can see a new.csv
file inresults
directory.
Name | MA-MRR | Our model |
---|---|---|
official run 8: epoch 80 Single_high | 0.62692 | No |
late submission epoch 100 Single_high | 0.63668 | Google Driver |
late submission epoch 100 multi_sorted | 0.64079 | same with last line |
@inproceedings{xu2022transfer,
title={Transfer learning with self-supervised vision transformer for large-scale plant identification},
author={Xu, Mingle and Yoon, Sook and Jeong, Yongchae and Lee, Jaesu and Park, Dong Sun},
booktitle={International conference of the cross-language evaluation forum for European languages (Springer;)},
pages={2253--2261},
year={2022}
}
@article{xu2022transfer,
title={Transfer learning for versatile plant disease recognition with limited data},
author={Xu, Mingle and Yoon, Sook and Jeong, Yongchae and Park, Dong Sun},
journal={Frontiers in Plant Science},
volume={13},
pages={4506},
year={2022},
publisher={Frontiers}
}
Our model is heavily based on MAE.