Deformable Convolution in 2D and 3D

PyTorch implementation of deformable convolution layers for 2D and 3D inputs with learnable offset fields.

Installation

From source

# Clone the repository
git clone https://github.com/momenator/deform_conv.git
cd deform_conv

# Install in development mode
pip install -e .

# Or install with development dependencies
pip install -e ".[dev]"

Requirements

• Python >= 3.8
• PyTorch >= 1.7.0
• NumPy >= 1.19.0

Usage

3D Deformable Convolution

import torch
from deformable_conv_3d import deform_conv3d

# Create input tensor (batch_size, channels, depth, height, width)
image = torch.rand(1, 16, 64, 64, 64)

# Create 3D deformable convolution layer
def_conv3 = deform_conv3d(in_c=16, out_c=32, kernel_size=(3, 3, 3))

# Forward pass
result = def_conv3(image)
print(result.shape)  # torch.Size([1, 32, 62, 62, 62])

2D Deformable Convolution

import torch
from deformable_conv_2d import ConvOffset2D

# Create input tensor (batch_size, channels, height, width)
image = torch.rand(1, 3, 32, 32)

# Create 2D deformable convolution layer
def_conv2 = ConvOffset2D(filters=3)

# Forward pass
result = def_conv2(image)
print(result.shape)  # torch.Size([1, 3, 32, 32])

Using the package import

import torch
from deform_conv import deform_conv3d, ConvOffset2D, ConvOffset3D

# Use as shown above

Development

Setup development environment

# Install with development dependencies
pip install -e ".[dev]"

# Run code formatter
black .

# Run linter
ruff check .

# Run type checker
mypy deformable_conv_2d.py deformable_conv_3d.py

Running tests

# Run all tests
pytest

# Run with coverage
pytest --cov=deform_conv --cov-report=html

License

MIT License - see LICENSE file for details.

Acknowledgments

Adapted from pytorch-deform-conv