Learning Neural Parametric 3D Breast Shape Models for Metrical Surface Reconstruction From Monocular RGB Videos
Maximilian Weiherer, Antonia von Riedheim, Vanessa Brébant, Bernhard Egger, Christoph Palm
arXiv 2025
Abstract
We present a neural parametric 3D breast shape model and, based on this model, introduce a low-cost and accessible 3D surface reconstruction pipeline capable of recovering accurate breast geometry from a monocular RGB video. In contrast to widely used, commercially available yet prohibitively expensive 3D breast scanning solutions and existing low-cost alternatives, our method requires neither specialized hardware nor proprietary software and can be used with any device that is able to record RGB videos. The key building blocks of our pipeline are a state-of-the-art, off-the-shelf Structure-from-motion pipeline, paired with a parametric breast model for robust and metrically correct surface reconstruction. Our model, similarly to the recently proposed implicit Regensburg Breast Shape Model (iRBSM), leverages implicit neural representations to model breast shapes. However, unlike the iRBSM, which employs a single global neural signed distance function (SDF), our approach---inspired by recent state-of-the-art face models---decomposes the implicit breast domain into multiple smaller regions, each represented by a local neural SDF anchored at anatomical landmark positions. When incorporated into our surface reconstruction pipeline, the proposed model, dubbed liRBSM (short for localized iRBSM), significantly outperforms the iRBSM in terms of reconstruction quality, yielding more detailed surface reconstruction than its global counterpart. Overall, we find that the introduced pipeline is able to recover high-quality 3D breast geometry within an error margin of less than 2 mm. Our method is fast (requires less than six minutes), fully transparent and open-source, and---together with the model---publicly available.
3D Breast Surface Reconstruction software
We provide an easy-to-use graphical user interface for our 3D reconstruction pipeline that runs on macOS (CPU-only) and Windows (CPU and NVIDIA graphics cards, if available). Just Drag&Drop an RGB video into our application, click six landmarks in a pre-selected frontal image, and hit the Process button.
How to take a video: To capture a video, please move in a 180-degree circular arc around the subject at an approximately constant speed (avoid standing still; keep moving smoothly around the subject), and make sure that the region between the belly button and sternal notch (both included) remains visible throughout the capture. Try to avoid using zoom while capturing, and ask the subject to hold still. Please also ensure a static background and good lighting conditions. A video should be about 10 to 20 seconds long.
Tips and tricks: In the default setting, 30 images will be extracted from the input video. If you want to change this number, adjust it under Settings. For a fast but less accurate reconstruction, as few as three images will work. To speed up processing even more, we recommend reducing the max. number of query points to the lowest setting (2048), and using a voxel resolution of 128. Settings can be changed anytime; just hit the Process button again to re-compute (if you don't change the SfM settings, this step will not be re-computed and the reconstruction becomes much faster). Right click on the reconstructed surface mesh to export it as PLY file.
System requirements and installation: The software requires about 15 GB of memory, and we recommend running the software on a computer with an NVIDIA graphics card (it's quite slow on the CPU). The very first start of the software can take a little longer -- this is completely normal and expected. To install under macOS, you'd need to remove the quarantine flag that is automatically added when downloading software from the internet which is not signed using an Apple Developer account. To do this, right after installation, open up a terminal and type xattr -cr "/Applications/3D Breast Surface Reconstruction.app". You need to do this only once! Please find more information here. We've tested the software on an Apple MacBook Pro M4 with the latest version of macOS Tahoe, and an Apple MacBook Pro M1 Max with macOS Sonoma 14.5. It also runs on a MacBook Pro with an Intel Quad-Core i7 processor and macOS Sequoia 15.6.1. We further tested on Windows 10 and Windows 11, with and without an NVIDIA graphics card.
If you encounter any problems with the software or just want to send some feedback, please reach out to Maximilian Weiherer.
Download the liRBSM and our software
A checkpoint of the trained model and our 3D Breast Surface Reconstruction software can be downloaded here. Please enter your full name, institution, and valid e-mail address to get access. Upon submission, an e-mail containing the download link will be sent to the address provided (please provide an institutional e-mail address if possible -- Gmail won't work!).
Citation
If you use the liRBSM or our 3D Breast Reconstruction software, please cite
@misc{weiherer2025lirbsm,
title={Learning Neural Parametric 3D Breast Shape Models for Metrical Surface Reconstruction From Monocular RGB Videos},
author={Weiherer, Maximilian and von Riedheim, Antonia and Brébant, Vanessa and Egger, Bernhard and Palm, Christoph},
archivePrefix={arXiv},
eprint={2510.13540},
year={2025}
}