KO | EN

GitLite

Minimalist Open Source Repository Search Engine

gitlite — search

#dsh #dsh-plugin #ai-agent #developer-tools #codex #typescript #cli #deepseek-harness #claude-code #ai-agents #mcp #rust

FocusMIL-and-other-max-pooling-methods

★ 10 Open GitHub ↗

Official code for "From Correlation to Causation: Max-Pooling-Based Multi-Instance Learning Leads to More Robust WSI Classification". FocusMIL, CausalMIL, mi-Net, and representative attention-based MIL baselines.

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# From Correlation to Causation:<br>Max-Pooling-Based Multi-Instance Learning Leads to<br>More Robust Whole Slide Image Classification

### FocusMIL and other awesome max-pooling methods

[Xin Liu](https://researchers.adelaide.edu.au/profile/xin.liu),
[Weijia Zhang](https://www.weijiazhangxh.com/),
[Wei Tang](https://tangw-seu.github.io/),
[Thuc Duy Le](https://researchers.adelaide.edu.au/profile/thuc.le),
[Jiuyong Li](https://researchers.adelaide.edu.au/profile/jiuyong.li),
[Lin Liu](https://researchers.adelaide.edu.au/profile/lin.liu),
[Min-Ling Zhang](https://palm.seu.edu.cn/zhangml/)

Adelaide University &nbsp;·&nbsp; Southeast University &nbsp;·&nbsp; University of Newcastle &nbsp;·&nbsp; MBZUAI

[![arXiv](https://img.shields.io/badge/arXiv-2408.09449-b31b1b.svg?style=for-the-badge&logo=arxiv&logoColor=white)](https://arxiv.org/abs/2408.09449)
[![HuggingFace Dataset](https://img.shields.io/badge/🤗%20Dataset-focusmil--camelyon16-ffce1c.svg?style=for-the-badge)](https://huggingface.co/datasets/Raymvp12/focusmil-camelyon16)

</div>

---

> **📦 Data release.** The Camelyon16 features and the **Standard MIL Test** set are
> now public on HuggingFace: **🤗 [Raymvp12/focusmil-camelyon16](https://huggingface.co/datasets/Raymvp12/focusmil-camelyon16)**
> (includes the files needed to compute the **FROC** tumor-localization metric).
> See [docs/DATASETS.md](docs/DATASETS.md) to download and set it up.

> The December 2025 version was submitted to CVPR 2026 (reviewer scores 5 / 4 / 3,
> recommended to Findings; we opted out). To guarantee the most rigorous reproduction
> and the fairest possible comparison across **all** baselines, we re-ran every
> experiment from scratch. You can refer to our results in [`results/`](results/).

## Introduction

Attention-based MIL and TransMIL-style methods actually **violate the standard MIL
assumption**. Because they pool instances through a *learnable weighted combination*
over all instances, they can exploit **negative evidence** — a pattern whose presence
correlates with the *negative* bag label. For some positive bags, when such
negative-correlated patterns appear they push the bag score **down**, and can even
wrongly flip a positive bag into a **false negative** — a missed diagnosis. This is
especially damaging when the positive rate is low, or the positive evidence is not
strong enough.

**Max-pooling** MIL structurally blocks this negative evidence: the bag score is
determined by the single instance with the highest positive-class score, so a
max-pooling model predicts the bag label **solely from the strength of positive
evidence** and cannot exploit negative-label-related evidence. This one-sided
prediction makes it especially **robust to the bias in negative bags**, at both
training and test time. However, max-pooling-based models have performed **poorly in
prior literature**. We find this is because they **mechanically memorize** the
association between positive-bag instances and their bag labels, overfitting the
training set.

This is the core idea of **FocusMIL**: a **variational information bottleneck (VIB)**
regularizes the latent space so the model *cannot* memorize the association between
features and labels. (We also tried dropout; dropout performs poorly on Camelyon17.)
We further add a **minibatch optimization**, which makes max-pooling training
noticeably more stable and thereby improves performance.

FocusMIL provides a **minimal yet effective instantiation of the max-pooling inductive
bias**. It can serve as a simple foundation for developing more sophisticated and powerful
max-pooling-based MIL methods. Beyond WSI classification, its VIB-based regularization may
also be applied to improve the training of max-pooling models in other MIL applications.
We welcome further extensions and applications of FocusMIL. If you have related ideas or
are interested in potential collaboration, please feel free to contact me.

## Methods (max-pooling)

The proposed / studied max-pooling models:

| Method | Camelyon16 | Camelyon17 |
|---|---|---|
| **FocusMIL** (ours) — per-instance VIB + max-pool | `train_ablation.py --variant focusmil_bs{N}_b{β}` | `train_c17_focusmil.py` |
| **CausalMIL** — an iVAE-based max-pooling MIL method (conditional KL & reconstruction loss applied only on the max-scoring instance) | `train_causalmil.py` | `train_c17_causalmil.py` |
| **mi-Net** (plain max-pool) | `train_ablation.py --variant maxmil_dp04_bs1` | `train_c17_maxmil.py` |

For each bag, the score is `max_i σ(f(z_i))` over instances.
`β = KL-coef / BCE-coef` weights the bottleneck; the default is **β = 1**.
You can also try **β = 0.1**.

Comparison methods (also reproduced here): DTFD-MIL, ACMIL, MHIM, IBMIL, AEM-MIL,
ABMIL, TransMIL, Conjunctive, DSMIL.

**Patch score for attention-based methods.** For the attention baselines we take the
attention network's output **logit** and apply a **sigmoid** as the per-patch score.
We tried several alternatives; this choice gives the highest patch-level and FROC metrics.

## Setup

```bash
pip install -r requirements.txt          # PyTorch + numpy/h5py/sklearn/skimage/...
# FROC (Camelyon16) also needs the OpenSlide C library:  apt-get install openslide-tools
```

Every on-disk path is configured through environment variables (see
[`mil_common/config.py`](mil_common/config.py)); nothing is hardcoded:

```bash
export CAMELYON16_ROOT=/path/to/camelyon16      # feature dirs + FROC features
export CAMELYON17_ROOT=/path/to/camelyon17      # feature .h5 files + camelyon17.csv
export C16_MASK_DIR=$CAMELYON16_ROOT/mask       # FROC only: extracted GT tumor masks
export MIL_CKPT_DIR=$PWD/checkpoints
export MIL_RESULTS_DIR=$PWD/results
export C16_FROC_OUTPUTS=$PWD/froc_outputs
```

## Data

- **Camelyon16** — ResNet18 (`c16_real`) and CTransPath (`c16_real_ctrans`) features, on
  **🤗 [Raymvp12/focusmil-camelyon16](https://huggingface.co/datasets/Raymvp12/focusmil-camelyon16)**.
- **Camelyon16 Standard MIL Test** (`c16_semi075`) — semi-synthetic robustness test with an
  inverted spurious correlation; same feature format and labels as `c16_real`. Same HF repo.
- **Camelyon17** — not part of our release; we use the features from the
  **[AEM repository](https://github.com/dazhangyu123/AEM)**.

See [docs/DATASETS.md](docs/DATASETS.md) for download and directory layout.

## Usage

```bash
# --- Camelyon16: proposed FocusMIL (ResNet18 backbone), seed 1 ---
cd camelyon16
python train_ablation.py --variant focusmil_bs3_b1 --data_tag c16_real --seed 1 --gpu 0

# a comparison method (mi-Net / plain max-pool):
python train_ablation.py --variant maxmil_dp04_bs1 --data_tag c16_real --seed 1 --gpu 0

# --- Camelyon16 FROC localization for a trained checkpoint ---
# --method must match the variant the checkpoint was trained with
# (the ckpt is <data_tag>__<method>_seed<seed>.pth)
python froc/extract_patch_csv.py --method focusmil_bs3_b1 --seed 1 --gpu 0
python froc/compute_froc.py --csv_dir $C16_FROC_OUTPUTS/focusmil_bs3_b1_seed1 \
       --out_json froc_focusmil_seed1.json --skip 021,114 --tolerance_um 150 --nms_radius_um 150

# --- Camelyon17: proposed FocusMIL (ResNet18), seed 1 ---
cd ../camelyon17
python train_c17_focusmil.py --batch_size 1 --instance_latent_dim 24 --seed 1 --gpu 0 --tag c17_focusmil_r18
```

Reproduce every method over 5 seeds:

```bash
bash camelyon16/scripts/run_all.sh                                     # ResNet18
DATA_TAG=c16_real_ctrans DFEAT=768 bash camelyon16/scripts/run_all.sh  # CTransPath
bash camelyon17/scripts/run_all.sh                                     # ResNet18
FEAT=pathgen IN_DIM=768 bash camelyon17/scripts/run_all.sh             # PathGen
```

`python aggregate.py --results_dir <dir>` collates a set of per-run JSONs into mean ± std.

### Camelyon16 Standard MIL Test

A controlled robustness test for the **standard MIL assumption**. It uses semi-synthetic
Camelyon16 features (same format and labels as `c16_real`) in which a *spurious,
non-causal* correlation is **inverted**. A method that exploits this **negative evidence**
is fooled and its slide AUC collapses, while a max-pooling model that predicts only from positive evidence stays robust. Point any
trainer at the `c16_semi075` data tag:

```bash
DATA_TAG=c16_semi075 bash camelyon16/scripts/run_all.sh     # all methods, 5 seeds
# single method:
python camelyon16/train_ablation.py --variant focusmil_bs3_b1 --data_tag c16_semi075 --seed 1 --gpu 0
```

Compare slide AUC on `c16_semi075` against `c16_real`: a large drop exposes a method that
relies on negative evidence and violates the standard MIL assumption.

## Contact

If you have any questions, or you feel any comparison method is implemented
inappropriately, please contact me at **xin.liu@adelaide.edu.au**.

## Citation

If you find this repository useful, please cite our paper:

```bibtex
@misc{liu2025correlationcausationmaxpoolingbasedmultiinstance,
      title={From Correlation to Causation: Max-Pooling-Based Multi-Instance Learning Leads to More Robust Whole Slide Image Classification},
      author={Xin Liu and Weijia Zhang and Wei Tang and Thuc Duy Le and Jiuyong Li and Lin Liu and Min-Ling Zhang},
      year={2025},
      eprint={2408.09449},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2408.09449},
}
```

## License

MIT — see [LICENSE](LICENSE).

FocusMIL-and-other-max-pooling-methods

Explore Similar Repositories

Was this content helpful?

Recent Feedback

Download README