MTLearn

MTLearn (Morphological Tree Learning) is a C++/Python research library for learnable connected operators based on morphological trees. The Python package is published as mtlearn.

The library explores a simple idea: connected morphology can become a structural prior for deep neural networks. Instead of processing images only through local pixel-wise operations, connected operators reason over components, regions, shape, contrast, and hierarchy. This makes them naturally interpretable and well-suited for tasks where structure matters.

Classical connected filters are powerful, but they usually depend on hard keep/discard decisions and manually selected attribute thresholds. This limits their integration into end-to-end trainable neural architectures.

MTLearn provides a stable implementation platform for this research direction. It currently includes Connected Filter Preprocessing (CFP), and is intended to grow toward trainable connected-operator layers, differentiable or learnable attribute criteria, self-dual tree representations, intermediate network insertions, and scalable implementations.

Main Features

• Connected Filter Preprocessing (CFP): the current main model, available as mtlearn.layers.ConnectedFilterPreprocessingLayer. CFP replaces hard connected-filter decisions with a differentiable sigmoid gate over normalized tree-node attributes.

• Stable morphology interface: mtlearn.morphology builds max-trees, min-trees, and tree-of-shapes through a backend-independent API.

• PyTorch integration: CFP layers are torch.nn.Module objects with trainable parameters, checkpoint helpers, and dataset-level normalization utilities.

• Validation utilities: includes C++ tests, Python tests, gradient checks, reference implementations, notebook utilities, and registered dataset download helpers.

Install

The Python package is available from PyPI as mtlearn:

pip install mtlearn

See the guides for installation and source builds:

• docs/installation.md
• docs/development.md

Quick Start

Create a CFP layer and run a forward pass:

import torch
from mtlearn import morphology
from mtlearn.layers import ConnectedFilterPreprocessingLayer

cfp_layer = ConnectedFilterPreprocessingLayer(
    in_channels=1,
    filter_specs=[
        {
            "tree_type": morphology.TreeType.MAX_TREE,
            "attributes": (
                morphology.AttributeType.AREA,
                morphology.AttributeType.CIRCULARITY,
            ),
        }
    ],
    device="cpu",
    scale_mode="none",
)

x = torch.tensor([[[[1, 2], [3, 4]]]], dtype=torch.float32)
y = cfp_layer(x)

assert y.shape == x.shape

For tree-of-shapes specs, normalization, caching, checkpoint helpers, and attribute selection, see the guides listed below.

Documentation

The main documentation is built with Sphinx from docs/source.

• Documentation home
• docs/installation.md
• docs/development.md
• Attribute guide
• CFP preparation and streaming statistics
• RRPR guide

Examples and Reproducibility

Executable examples are available in notebooks/. Notebook dependency setup is documented in docs/installation.md.

For ICPR 2026 Reproducible Research in Pattern Recognition (RRPR) review, use RRPR guide as the canonical guide for setup, datasets, notebook execution, hardware notes, and current reproducibility scope.

Scope and API Boundaries

MTLearn is a research-oriented library. CFP is the current validated trainable connected-operator layer, with support for max-tree, min-tree, and tree-of-shapes workflows; mixed tree specs; multi-attribute dataset-level normalization; cached preprocessing; and PyTorch forward/backward for CFP parameters on CPU or CUDA tensors.

User code should interact with morphology through the public Python facade mtlearn.morphology and the primary CFP layer in mtlearn.layers, rather than depending on backend-specific APIs. The backend is mmcfilters, but the top-level Python package mmcfilters is not required as a runtime dependency of mtlearn.

Citation

If you use the CFP layer in your work, please cite:

Wonder A. L. Alves, Lucas de P. O. Santos, Ronaldo F. Hashimoto, Nicolas Passat, Anderson H. R. Souza, Dennis J. Silva, Yukiko Kenmochi. A trainable connected filter preprocessing layer based on component trees. International Conference on Pattern Recognition (ICPR), 2026, Lyon, France. ⟨hal-05575141⟩