Model Card for PathDINOv3

PathDINOv3 is a powerful self-supervised foundation model strictly designed for advanced pathology image analysis, trained from scratch utilizing the cutting-edge DINOv3 framework.

The model relies on a Vision Transformer architecture combined with DINOv3 self-supervised pre-training to establish robust and highly discriminative token-level representations across varied tissue morphologies.

---

Using PathDINOv3 to extract features from pathology images

import timm
import torch
import torchvision.transforms as transforms

model = timm.create_model('hf_hub:minxoy/PathDINOv3', pretrained=True, init_values=1e-5, dynamic_img_size=True)

preprocess = transforms.Compose([
            transforms.Resize(224),
            transforms.ToTensor(),
            transforms.Normalize(mean=(0.485, 0.456, 0.406), std=(0.229, 0.224, 0.225)),])

model = model.to('cuda')
model.eval()

input = torch.randn([1, 3, 224, 224]).cuda()

with torch.no_grad():
    output = model(input)

Training Pipeline

Self Supervised Learning: https://github.com/facebookresearch/dinov3