---
tags:
- image-classification
- birder
library_name: birder
license: apache-2.0
---

# Model Card for davit_tiny_il-all

A Dual Attention Vision Transformer (DaViT) image classification model. This model was trained on the `il-all` dataset (all the relevant bird species found in Israel inc. rarities).

The specie list is based on data from <https://www.israbirding.com/checklist/>.

## Model Details

- **Model Type:** Image classification / detection backbone
- **Model Stats:**
  - Params (M): 28.0
  - Image size: 384 x 384
- **Dataset:** il-all (550 classes)

- **Papers:**
  - DaViT: Dual Attention Vision Transformers: <https://arxiv.org/abs/2204.03645>

## Model Usage

### Image Classification

```python
import birder
from birder.inference.classification import infer_image

(net, class_to_idx, signature, rgb_stats) = birder.load_pretrained_model("davit_tiny_il-all", inference=True)

# Get the image size the model was trained on
size = birder.get_size_from_signature(signature)

# Create an inference transform
transform = birder.classification_transform(size, rgb_stats)

image = "path/to/image.jpeg"  # or a PIL image
(out, _) = infer_image(net, image, transform)
# out is a NumPy array with shape of (1, num_classes)
```

### Image Embeddings

```python
import birder
from birder.inference.classification import infer_image

(net, class_to_idx, signature, rgb_stats) = birder.load_pretrained_model("davit_tiny_il-all", inference=True)

# Get the image size the model was trained on
size = birder.get_size_from_signature(signature)

# Create an inference transform
transform = birder.classification_transform(size, rgb_stats)

image = "path/to/image.jpeg"  # or a PIL image
(out, embedding) = infer_image(net, image, transform, return_embedding=True)
# embedding is a NumPy array with shape of (1, embedding_size)
```

### Detection Feature Map

```python
from PIL import Image
import birder

(net, class_to_idx, signature, rgb_stats) = birder.load_pretrained_model("davit_tiny_il-all", inference=True)

# Get the image size the model was trained on
size = birder.get_size_from_signature(signature)

# Create an inference transform
transform = birder.classification_transform(size, rgb_stats)

image = Image.open("path/to/image.jpeg")
features = net.detection_features(transform(image).unsqueeze(0))
# features is a dict (stage name -> torch.Tensor)
print([(k, v.size()) for k, v in features.items()])
# Output example:
# [('stage1', torch.Size([1, 96, 96, 96])),
#  ('stage2', torch.Size([1, 192, 48, 48])),
#  ('stage3', torch.Size([1, 384, 24, 24])),
#  ('stage4', torch.Size([1, 768, 12, 12]))]
```

## Citation

```bibtex
@misc{ding2022davitdualattentionvision,
      title={DaViT: Dual Attention Vision Transformers},
      author={Mingyu Ding and Bin Xiao and Noel Codella and Ping Luo and Jingdong Wang and Lu Yuan},
      year={2022},
      eprint={2204.03645},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2204.03645}
}
```