Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

DaCapo: a modular deep learning framework for scalable 3D image segmentation

William Patton, Jeff L. Rhoades, Marwan Zouinkhi, David G. Ackerman, Caroline Malin-Mayor, Diane Adjavon, Larissa Heinrich, Davis Bennett, Yurii Zubov, CellMap Project Team, Aubrey V. Weigel, Jan Funke

TL;DR

DaCapo is introduced, highlighting its modular structure, efficient experiment management tools, and scalable deployment capabilities, and its potential to improve access to large-scale, isotropic image segmentation is discussed.

Abstract

DaCapo is a specialized deep learning library tailored to expedite the training and application of existing machine learning approaches on large, near-isotropic image data. In this correspondence, we introduce DaCapo's unique features optimized for this specific domain, highlighting its modular structure, efficient experiment management tools, and scalable deployment capabilities. We discuss its potential to improve access to large-scale, isotropic image segmentation and invite the community to explore and contribute to this open-source initiative.

DaCapo: a modular deep learning framework for scalable 3D image segmentation

TL;DR

DaCapo is introduced, highlighting its modular structure, efficient experiment management tools, and scalable deployment capabilities, and its potential to improve access to large-scale, isotropic image segmentation is discussed.

Abstract

DaCapo is a specialized deep learning library tailored to expedite the training and application of existing machine learning approaches on large, near-isotropic image data. In this correspondence, we introduce DaCapo's unique features optimized for this specific domain, highlighting its modular structure, efficient experiment management tools, and scalable deployment capabilities. We discuss its potential to improve access to large-scale, isotropic image segmentation and invite the community to explore and contribute to this open-source initiative.
Paper Structure (8 sections, 1 figure)

This paper contains 8 sections, 1 figure.

Table of Contents

  1. Introduction
  2. Training Setup
  3. Task Specification
  4. Model Architecture
  5. Blockwise Inference & Post-processing
  6. Compute Contexts
  7. Conclusion
  8. Bibliography

Figures (1)

  • Figure 1: Anatomy of DaCapo.a) DaCapo framework is subdivided into individually configurable submodules. For model training, DaCapo takes information about data sources and configuration of training hyperparameters, such as the number of iterations or batch size (see panel b for more details). Trained models can then be quickly applied to new datasets. Both training and post-training tasks can be run in various compute contexts, utilizing the power of cloud and cluster resources, as well as the simplicity of local setups. b) DaCapo train requires specification of a datasplit, trainer, model, and task. The dataplit specifies which datasources should be used for training vs. validation scores. This can be supplied with a simple CSV file. Training hyperparameters such as number of training iterations, data augmentations, etc. are also easily configurable in the trainer. The network architecture to be used (model) and target representation the model is being trained to predict (task) are also specified here. c) DaCapo Apply is used after training is complete. Prediction and post-processing can be seamlessly scaled with DaCapo via blockwise processing of any size volume.