🐔 Chicks4FreeID

The very first publicly available chicken re-identification dataset is available on 🤗 Hugging Face: huggingface.co/datasets/dariakern/Chicks4FreeID

🤗 Usage

pip install datasets

Load the data:

from datasets import load_dataset
train_ds = load_dataset("dariakern/Chicks4FreeID", split="train")
train_ds[0]

Output:

{'crop': <PIL.PngImagePlugin.PngImageFile image mode=RGB size=2630x2630 at 0x7AA95E7D1720>,
'identity': 43}

Tip

Find more information on how to work with 🤗 huggingface.co/docs/datasets

📊 Baseline

To establish a baseline on the dataset, we explore 3 approaches

1. We evaluate the SotA model in animal re-identification: MegaDescriptor-L-384, a feature extractor, pre-trained on many species and identities.

   timm.create_model("hf-hub:BVRA/MegaDescriptor-L-384", pretrained=True)

2. We train MegaDescriptor-L-384's underlying architecture; a Swin-Transformer, in the same way it has been used to build the MegaDescriptor-L-384, but now on our own dataset.

   timm.create_model('swin_large_patch4_window12_384')

3. We train a Vision Transformer (ViT-B/16) as a fully supervised baseline, and focus on embeddings by replacing the classifier head with a linear layer.

   from torchvision.models import vit_b_16

Evaluation settings are based on:

• Linear: SimCLR
• k-NN: InstDist

Metrics are from torchmetrics

• mAP: MulticlassAveragePrecision(average="macro")
• top1: MulticlassAccuracy(top_k=1)
• top5: MulticlassAccuracy(top_k=5)

Below are the metrics for the test set. Standard deviations are based on 3 runs:

Setting	Evaluation	mAP	top-1	top-5
MegaDescriptor-L-384 (frozen)	k-NN	0.649 ± 0.044	0.709 ± 0.026	0.924 ± 0.027
MegaDescriptor-L-384 (frozen)	Linear	0.935 ± 0.005	0.883 ± 0.009	0.985 ± 0.003
Swin-L-384	k-NN	0.837 ± 0.062	0.881 ± 0.041	0.983 ± 0.010
Swin-L-384	Linear	0.963 ± 0.022	0.922 ± 0.042	0.987 ± 0.012
ViT-B/16	k-NN	0.893 ± 0.010	0.923 ± 0.005	0.985 ± 0.019
ViT-B/16	Linear	0.976 ± 0.007	0.928 ± 0.002	0.990 ± 0.012

The most interesting observation in this table is that, even though the MegaDescriptor-L-384 feature extractor has never seen our dataset, its embeddings are still relatively helpful in identifiying the chickens, even when compared to the fully supervised approaches.

🧑‍💻 Replicate the baseline

git clone https://github.com/DariaKern/Chicks4FreeID
cd Chicks4FreeID
pip install requirements.txt
python run_baseline.py

You can pass different options, depending on your hardware configuration

python run_baseline.py --devices=4 --batch-size-per-device=128 

For a full list of arguments type

python run_baseline.py --help

In a sepearte shell, open tensorboard to view progress and results

tensorboard --logdir baseline_logs

Note

Differnt low-level accelerator implementations (TPU, MPS, CUDA) yield different results. The original hardware config for the reported results is based on the MPS implementation accessible on a 64GB Apple M3 Max chip (2023) 💻 - it is recommened to run the baseline script with at least 64GB of VRAM / Shared RAM. On this device, one run takes around 9:30h

⏳ Timeline

• [2024/05/30] DOI created: https://doi.org/10.57967/hf/2345
• [2024/05/23] the first version of the dataset was uploaded to Hugging Face. https://huggingface.co/datasets/dariakern/Chicks4FreeID

📝 Papers and systems citing the Chicks4FreeID dataset

coming soon ...

🖋️ Citation

@misc{kern2024Chicks4FreeID,
      title={Chicks4freeID}, 
      author={Daria Kern and Tobias Schiele and Ulrich Klauck and Winfred Ingabire},
      year={2024},
      doi={https://doi.org/10.57967/hf/2345},
      note={under review}
}