from PIL import Image
import torch
import timm
from timm.data import create_transform
from timm.data import resolve_model_data_config

model = timm.create_model("hf-hub:hassonofer/vit_so150m_patch14_reg4_biodino_252", pretrained=True)
model.eval()

data_config = resolve_model_data_config(model)
transform = create_transform(**data_config, is_training=False)

image = Image.open("path/to/image.jpg").convert("RGB")
x = transform(image).unsqueeze(0)

with torch.inference_mode():
    embedding = model(x)

print(embedding.shape)  # torch.Size([1, 896])

Relationship To The Original Release

This repository is only the TIMM-compatible 252px checkpoint. The original Birder release includes the Birder-native checkpoints and additional documentation for the 224px, 252px and 336px variants.

Please cite or link the original Birder release when using this TIMM conversion.