MLforHealthcare/dinov2-nuclei-segmentation

DINOv2 Nuclei Segmentation Model

Model Details

• Format: PyTorch .pth checkpoint
• Architecture: DINOv2-Base (vitb14) with custom decoder
• Training Data: PanNuke dataset
• Input Size: 252x252
• Classes: Background, Neoplastic, Inflammatory, Connective, Dead, Epithelial

Usage

from huggingface_hub import hf_hub_download

model = DinoV2ForSegmentation(num_classes=6)
model.load_state_dict(
    torch.load(
        hf_hub_download(
            repo_id="{model_repo}",
            filename="{best_checkpoint}"
        )
    )
)