News

Metrics slightly updated! Retrained with PyTorch NGC Docker Container 20.07 and on Ampere GPUs only (3080 / 3090)
EXIF metadata available! You can read it dirrectly from the images.

Danish Fungi 2020 - Not Just Another Image Recognition Dataset

Introduction

Supplementary material to: Danish Fungi 2020 - Not Just Another Image Recognition Dataset

In order to support research in fine-grained plant classification and to allow full reproducibility of our results, we share the Training Logs and Trained scripts.

The Images, Checkpoints and Metadata are not included based on size constrains and will be published after the review.

Training Data

Available at -> https://sites.google.com/view/danish-fungi-dataset

Training

Download PyTorch NGC Docker Image and RUN docker container

docker pull nvcr.io/nvidia/pytorch:21.07-py3
docker run --gpus all -it --rm -v local_dir:container_dir nvcr.io/nvidia/pytorch:21.07-py3

Install dependencies inside docker container

pip install pandas seaborn timm albumentation tqdm efficientnet_pytorch pretrainedmodels

RUN jupyterlab and start training / experiments

jupyter lab --ip 0.0.0.0 --port 8888 --allow-root

Check your paths!

Results

CNN Performance Evaluation

Classification performance of selected CNN architectures on DF20 and DF20 - Mini. All networks share the settings described in Section 6.1 and were trained on 299×299 images.

	Top1 [%]	Top3 [%]	F1	Top1 [%]	Top3 [%]	F1
MobileNet-V2	65.58	83.65	0.559	69.77	85.01	0.606
ResNet-18	62.91	81.65	0.514	67.13	82.65	0.580
ResNet-34	65.63	83.52	0.559	69.81	84.76	0.600
ResNet-50	68.39	85.22	0.587	73.49	87.13	0.649
EfficientNet-B0	67.94	85.71	0.567	73.65	87.55	0.653
EfficientNet-B1	68.35	84.67	0.572	74.08	87.68	0.654
EfficientNet-B3	69.59	85.55	0.590	75.69	88.72	0.673
EfficientNet-B5	68.76	85.00	0.579	76.10	88.85	0.678
Inception-V3	65.91	82.97	0.535	72.10	86.58	0.630
InceptionResnet-V2	64.67	81.42	0.542	74.01	87.49	0.651
Inception-V4	67.45	82.78	0.560	73.00	86.87	0.637
SE-ResNeXt-101-32x4d	72.23	87.28	0.620	77.13	89.48	0.693
----------------	----	----	----	----	----	----
Dataset	DF20M	DF20M	DF20M	DF20	DF20	DF20

ViT x CNN Performance Evaluation

Classification results of selected CNN and ViT architectures on DF20 and DF20,-,Mini dataset for two input resolutions [299𐄂299, 384𐄂384].

	Top1 [%]	Top3 [%]	F1	Top1 [%]	Top3 [%]	F1
EfficientNet-B0	65.66	83.35	0.531	70.33	85.19	0.613
EfficientNet-B3	67.39	83.74	0.550	72.51	86.77	0.634
SE-ResNeXt-101	68.87	85.14	0.585	74.26	87.78	0.660
ViT-Base/16	70.11	86.81	0.600	73.51	87.55	0.655
ViT-Large/16	71.04	86.15	0.603	75.29	88.34	0.675
----------------	----	----	----	----	----	----
Dataset	DF20M	DF20M	DF20M	DF20	DF20	DF20

	Top1 [%]	Top3 [%]	F1	Top1 [%]	Top3 [%]	F1
EfficientNet-B0	69.62	85.96	0.582	75.35	88.67	0.670
EfficientNet-B3	71.59	87.39	0.613	77.59	90.07	0.699
SE-ResNeXt-101	74.23	88.27	0.651	78.72	90.54	0.708
ViT-Base/16	74.23	89.12	0.639	79.48	90.95	0.727
ViT-Large/16	75.85	89.95	0.669	80.45	91.68	0.743
----------------	----	----	----	----	----	----
Dataset	DF20M	DF20M	DF20M	DF20	DF20	DF20

Metadata Usage Experiment

Performance gains from Fungus observation metadata: H - Habitat, S - Substrate, M - Month, and their combinations, on DF20.

DF20 - ViT-Large/16 with image size 384𐄂384.

H	M	S	Top1 [%]	Top3 [%]	F1
𐄂	𐄂	𐄂	80.45	91.68	0.743
✔	𐄂	𐄂	+1.50	+1.00	+0.027
𐄂	✔	𐄂	+0.95	+0.62	+0.014
𐄂	𐄂	✔	+1.13	+0.69	+0.020
𐄂	✔	✔	+1.93	+1.27	+0.032
✔	𐄂	✔	+2.48	+1.66	+0.044
✔	✔	𐄂	+2.31	+1.48	+0.040
✔	✔	✔	+2.95	+1.92	+0.053

DF20-Mini - ViT-Base/16 with image size 224𐄂224.

H	M	S	Top1	Top3	F1
𐄂	𐄂	𐄂	73.51	87.55	0.655
✔	𐄂	𐄂	+1.94	+1.50	+0.040
𐄂	✔	𐄂	+1.23	+0.95	+0.020
𐄂	𐄂	✔	+1.39	+1.17	+0.025
𐄂	✔	✔	+2.47	+1.98	+0.042
✔	𐄂	✔	+3.23	+2.47	+0.062
✔	✔	𐄂	+3.11	+2.30	+0.057
✔	✔	✔	+3.81	+2.84	+0.070

License

The code and dataset is released under the BSD License. There is some limitations for commercial usage. In other words, the training data, metadata, and models are available only for non-commercial research purposes only.

Citation

If you use Danish Fungi for your research or aplication, please consider citation:

@article{picek2021danish,
title={Danish Fungi 2020 - Not Just Another Image Recognition Dataset},
author={Lukáš Picek and Milan Šulc and Jiří Matas and Jacob Heilmann-Clausen and Thomas S. Jeppesen and Thomas Læssøe and Tobias Frøslev},
year={2021},
eprint={2103.10107},
archivePrefix={arXiv},
primaryClass={cs.CV}
}

Contact

[Lukas Picek](lukaspicek@gmail.com, picekl@ntis.zcu.cz)

Woodman718/DanishFungiDataset