ALICE

Model weights | Preprint paper

Introduction

ALICE (Agglomerative Learning via Integrated Computational pathology Embedding) is a unified general-purpose pathology foundation model trained through multi-stage agglomerative distillation, which sequentially distills eight vision-only, vision-language, and slide-level teacher models into dedicated modules of a single backbone. It provides a unified representation for ROI tissue analysis, vision-language multimodal understanding, and whole-slide clinical assessment, achieving the best average rank among task-matched pathology foundation models across 21 task scenarios, 96 downstream tasks, and 48 data sources.

The ALICE model includes three types of inference:

• vision_stage: extracts vision-only patch features from pathology image tensors.
• vl_stage: maps ALICE vision features to vision-language teacher-head features.
• slide_stage: aggregates patch-level ALICE features into slide-level features.

Model weights and Hugging Face AutoModel loading code are hosted at:

Quick Start

Installation

pip install torch torchvision timm transformers pillow huggingface_hub

Log in first:

huggingface-cli login

Using ALICE to extract tile-level image features, language-aligned image features, and whole-slide image features:

import torch
from PIL import Image

example_image_path = './example.jpg'

with torch.no_grad():
    # Vision-only: image -> raw vision feature
    # [B, 3, H, W] -> [B, 3840]
    image = Image.open(example_image_path).convert("RGB")
    image_tensor = alice.image_transform(image).unsqueeze(0)  # [1, 3, 224, 224]
    vision_features = alice.vision_stage(image_tensor)
    # vision_features: [B, 3840]

    # Vision-language: vision_stage output -> dict of teacher-head features
    # [B, 3840] -> dict[str, Tensor]
    vl_features = alice.vl_stage(vision_features)
    # vl_features["keep"]:     KEEP head,     [B, 768]
    # vl_features["conch_v1"]: CONCH v1 head, [B, 512]
    # vl_features["musk"]:     MUSK head,     [B, 1024]

    # Slide-level: patch_features + coords -> slide feature
    # [N, 3840] + [N, 2] -> [B, 2048]
    patch_features = torch.randn(100, 3840)
    coords = torch.randint(0, 10000, (100, 2))
    slide_features = alice.slide_stage(patch_features, coords=coords, patch_size_lv0=512)
    # slide_features: [B, 2048]

Image preprocessing can also use the following methods:

from torchvision import transforms
from torchvision.transforms import InterpolationMode

image_preprocess = transforms.Compose([
    transforms.Resize(224, interpolation=InterpolationMode.BICUBIC, antialias=True),
    transforms.CenterCrop(224),
    transforms.ToTensor(),
    transforms.Normalize(
        mean=(0.48145466, 0.4578275, 0.40821073),
        std=(0.26862954, 0.26130258, 0.27577711),
    ),
])

Output Summary

Stage	Input	Output
vision_stage	image tensor [B, 3, H, W]	feature tensor [B, 3840]
vl_stage	vision_stage output	dict with keep, conch_v1, musk features
slide_stage	patch features [N, 3840] or [B, N, 3840]	slide feature [B, 2048]

License

CC-BY-NC-ND-4.0

Citation

If ALICE is helpful to your research, please consider citing our work:

@misc{li2026alice,
      title={ALICE: Learning a General-Purpose Pathology Foundation Model from Vision, Vision-Language, and Slide-Level Experts}, 
      author={Jiawen Li and Tian Guan and Huijuan Shi and Xitong Ling and Mingxi Fu and Anjia Han and Chao He and Yonghong He},
      year={2026},
      eprint={2607.09526},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2607.09526}, 
}