Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Scale-Aware Self-Supervised Learning for Segmentation of Small and Sparse Structures

Jorge Quesada, Ghassan AlRegib

TL;DR

This work addresses the scale sensitivity of self-supervised learning for segmentation, showing that traditional SSL biases toward large structures underperform on small or sparse targets. It introduces a scale-aware augmentation by embedding small-window cropping into pretraining, compatible with any SSL loss, and validates the approach on seismic fault segmentation and neuroimaging cell delineation. Across these domains, the method yields notable gains for small targets (e.g., up to 13% in fault accuracy and 5% in cell delineation) while offering limited benefits for large structures, highlighting a clear scale-dependent effect. The findings advocate aligning SSL design with the intrinsic geometry of the downstream task and motivate adaptive multi-scale strategies for broad scientific imaging applicability.

Abstract

Self-supervised learning (SSL) has emerged as a powerful strategy for representation learning under limited annotation regimes, yet its effectiveness remains highly sensitive to many factors, especially the nature of the target task. In segmentation, existing pipelines are typically tuned to large, homogeneous regions, but their performance drops when objects are small, sparse, or locally irregular. In this work, we propose a scale-aware SSL adaptation that integrates small-window cropping into the augmentation pipeline, zooming in on fine-scale structures during pretraining. We evaluate this approach across two domains with markedly different data modalities: seismic imaging, where the goal is to segment sparse faults, and neuroimaging, where the task is to delineate small cellular structures. In both settings, our method yields consistent improvements over standard and state-of-the-art baselines under label constraints, improving accuracy by up to 13% for fault segmentation and 5% for cell delineation. In contrast, large-scale features such as seismic facies or tissue regions see little benefit, underscoring that the value of SSL depends critically on the scale of the target objects. Our findings highlight the need to align SSL design with object size and sparsity, offering a general principle for buil ding more effective representation learning pipelines across scientific imaging domains.

Scale-Aware Self-Supervised Learning for Segmentation of Small and Sparse Structures

TL;DR

This work addresses the scale sensitivity of self-supervised learning for segmentation, showing that traditional SSL biases toward large structures underperform on small or sparse targets. It introduces a scale-aware augmentation by embedding small-window cropping into pretraining, compatible with any SSL loss, and validates the approach on seismic fault segmentation and neuroimaging cell delineation. Across these domains, the method yields notable gains for small targets (e.g., up to 13% in fault accuracy and 5% in cell delineation) while offering limited benefits for large structures, highlighting a clear scale-dependent effect. The findings advocate aligning SSL design with the intrinsic geometry of the downstream task and motivate adaptive multi-scale strategies for broad scientific imaging applicability.

Abstract

Self-supervised learning (SSL) has emerged as a powerful strategy for representation learning under limited annotation regimes, yet its effectiveness remains highly sensitive to many factors, especially the nature of the target task. In segmentation, existing pipelines are typically tuned to large, homogeneous regions, but their performance drops when objects are small, sparse, or locally irregular. In this work, we propose a scale-aware SSL adaptation that integrates small-window cropping into the augmentation pipeline, zooming in on fine-scale structures during pretraining. We evaluate this approach across two domains with markedly different data modalities: seismic imaging, where the goal is to segment sparse faults, and neuroimaging, where the task is to delineate small cellular structures. In both settings, our method yields consistent improvements over standard and state-of-the-art baselines under label constraints, improving accuracy by up to 13% for fault segmentation and 5% for cell delineation. In contrast, large-scale features such as seismic facies or tissue regions see little benefit, underscoring that the value of SSL depends critically on the scale of the target objects. Our findings highlight the need to align SSL design with object size and sparsity, offering a general principle for buil ding more effective representation learning pipelines across scientific imaging domains.
Paper Structure (10 sections, 3 equations, 2 figures, 2 tables)

This paper contains 10 sections, 3 equations, 2 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Methodology
  4. Self-Supervised Pretraining
  5. Scale-Aware View Sampling
  6. Downstream Segmentation
  7. Results
  8. Experimental Setup
  9. Results and analysis
  10. Conclusion

Figures (2)

  • Figure 1: Differences in SSL segmentation depend on the nature of the target. Left: Seismic segmentation task with geological facies (large) and seismic faults (small). Right: Neuroimaging segmentation task from a mouse brain cutout, with axons (large) and cell bodies and vessels (small).
  • Figure 2: Qualitative segmentation results for CRACKS (left) and MTNeuro (right).