# Bowerbird Individual Classifier (ResNet50)

This repository contains the weights for a ResNet50-based individual ID classifier
trained on a set of Spotted Bowerbird individuals, as part of a project of the Fusani Lab
(see https://github.com/sarequi/Bowerbird-ID).

- Base model: `torchvision.models.resnet50` with `ResNet50_Weights.DEFAULT` (ImageNet)
- Input size: 512 × 512 RGB
- Normalization:
  - mean = [0.485, 0.456, 0.406]
  - std  = [0.229, 0.224, 0.225]
- Checkpoint file: `Bbird_individual_classifier.pth`

> This model is **not** generic. It is specific to the 16 individuals it was trained on.
> `NUM_CLASSES` must match the number of bird IDs used during training, unless the model is re-trained.

## Usage

```python
import torch
from torchvision.models import resnet50, ResNet50_Weights
from huggingface_hub import hf_hub_download

repo_id = "sarequi/bowerbird-individual-classifier"

# Download checkpoint
ckpt_path = hf_hub_download(
    repo_id=repo_id,
    filename="Bbird_individual_classifier.pth",
)

NUM_CLASSES = 16  # number of individuals used during training

# Rebuild model architecture
model = resnet50(weights=ResNet50_Weights.DEFAULT)
num_ftrs = model.fc.in_features
model.fc = torch.nn.Linear(num_ftrs, NUM_CLASSES)

# Load weights
state_dict = torch.load(ckpt_path, map_location="cpu")
model.load_state_dict(state_dict)
model.eval()