TACoS: Weakly Supervised Learning of Two-Dimensional Materials from Scribble Annotations to Precise Segmentation

2026-07-08 · Source: Takara TLDR - Daily AI Papers · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Physical Sciences & Chemistry · Depth: Expert, medium

Summary

TACoS is a specialized scribble segmentation framework designed for the precise pixel-level localization of two-dimensional material flakes, crucial for high-throughput screening. Addressing the limitations of costly dense annotations required by fully supervised methods, TACoS employs a weakly supervised approach. The framework integrates semi-supervised consistency learning with structured tree energy constraints, featuring an unlabeled weak-strong distribution alignment module that uses cosine consistency and a tree energy regularization module that generates structure-aware soft pseudo labels via minimum spanning trees. Additionally, it incorporates asymmetric regional contrast learning, which fuses high-confidence predictions with scribbles to form augmented labels and constructs category prototypes. This strategy enhances intra-class cohesion and inter-class separation, particularly in low-contrast edges and complex backgrounds. Experiments on graphene and MoS2 datasets show TACoS achieves over 96% of fully supervised performance with less than 0.6% annotated data, offering an efficient and scalable solution with superior structural coherence and boundary stability.

Key takeaway

For Machine Learning Engineers developing segmentation models for 2D materials, if you face high annotation costs, consider weakly supervised frameworks like TACoS. This method achieves over 96% of fully supervised performance with less than 0.6% annotated data. Implementing such approaches can significantly accelerate your model deployment. It also reduces your team's labeling burden for high-throughput screening applications.

Key insights

Weakly supervised learning with scribble annotations enables precise 2D material segmentation with minimal data.

Principles

• Consistency learning improves prediction alignment across augmented views.
• Tree energy constraints establish pixel affinity for structure-aware pseudo labels.
• Regional contrast learning enhances class separation in complex backgrounds.

Method

A unified framework integrates semi-supervised consistency learning with structured tree energy constraints, using cosine consistency and minimum spanning trees. Asymmetric regional contrast learning fuses high-confidence predictions with scribbles and applies contrastive constraints on challenging boundary pixels.

In practice

• Reduce annotation costs for material flake segmentation using scribbles.
• Improve segmentation accuracy in low-contrast images via regional contrast.

Topics

• Weakly Supervised Learning
• Image Segmentation
• 2D Materials
• Scribble Annotations
• Consistency Learning
• Contrastive Learning
• High-Throughput Screening

Code references

• liuxuexun/DBGroup
• krishnabits001/domain_specific_cl

Best for: AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• TACoS: Weakly Supervised Learning of Two-Dimensional Materials from Scribble Annotations to Precise Segmentation — TACoS enables precise 2D material segmentation using minimal scribble annotations via a novel semi-supervised framework.
• Learning Structured Visual Compositional Representations for Weakly Supervised Referring Expression Comprehension — Explicitly modeling structured visual representations, including unary and pairwise embeddings, significantly enhances weakly supervised referring expression comprehension.
• Bulk Labelling and Prodigy — Bulk labeling uses language model embeddings and dimensionality reduction to visually cluster and rapidly pre-label text subsets for annotation.
• Learning to Label: A Reinforced Self-Evolving Framework for Semi-supervised Referring Expression Segmentation — A reinforced self-evolving framework improves semi-supervised referring expression segmentation by learning to construct reliable pseudo-labels.
• Complex Layout Classification in the Wild: A Low-Resource Approach with Layout-Preserving Augmentations — Domain-aware augmentations significantly improve low-resource complex layout classification by focusing on global geometric arrangements.
• SegMix:Shuffle-based Feedback Learning for Semantic Segmentation of Pathology Images — Shuffle-based feedback learning improves weakly supervised semantic segmentation in pathology by adapting to multi-scale image characteristics.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Takara TLDR - Daily AI Papers.