Best Practices in Active Learning for Semantic Segmentation

Active learning is particularly of interest for semantic segmentation, where annotations are costly. Previous academic studies focused on datasets that are already very diverse and where the model is trained in a supervised manner with a large annotation budget. In contrast, data collected in many driving scenarios is highly redundant, and most medical applications are subject to very constrained annotation budgets. This work investigates the various types of existing active learning methods for semantic segmentation under diverse conditions across three dimensions - data distribution w.r.t. different redundancy levels, integration of semi-supervised learning, and different labeling budgets. We find that these three underlying factors are decisive for the selection of the best active learning approach. As an outcome of our study, we provide a comprehensive usage guide to obtain the best performance for each case. We also propose an exemplary evaluation task for driving scenarios, where data has high redundancy, to showcase the practical implications of our research findings.

• Publication year

  2023

• Venue

  DAGM

• Publication date

  2023-02-08

• Fields of study

  Medicine, Computer Science

• Identifiers
  DOI 10.1007/978-3-031-54605-1_28arXiv 2302.04075
• External record

  Open on Semantic Scholar

• Source metadata

  Semantic Scholar

• No claims are published for this paper.

• No concepts are published for this paper.

• Semantic Segmentation with Active Semi-Supervised Learning

  2022cited by this paper
• Semi-Supervised Semantic Segmentation Using Unreliable Pseudo-Labels

  2022cited by this paper
• All you need are a few pixels: semantic segmentation with PixelPick

  2021cited by this paper
• Semi-Supervised Active Learning with Temporal Output Discrepancy

  2021cited by this paper
• Revisiting Superpixels for Active Learning in Semantic Segmentation with Realistic Annotation Costs

  2021cited by this paper
• Classification of acoustical signals by combining active learning strategies with semi-supervised learning schemes

  2021cited by this paper
• Towards Robust and Reproducible Active Learning using Neural Networks

  2020cited by this paper
• GLISTER: Generalization based Data Subset Selection for Efficient and Robust Learning

  2020influential reference
• Dual Adversarial Network for Deep Active Learning

  2020cited by this paper
• Importance of Self-Consistency in Active Learning for Semantic Segmentation

  2020influential reference
• A2D2: Audi Autonomous Driving Dataset

  2020influential reference
• Task-Aware Variational Adversarial Active Learning

  2020cited by this paper
• Consistency-Based Semi-Supervised Active Learning: Towards Minimizing Labeling Cost

  2019cited by this paper
• Diverse mini-batch Active Learning

  2019influential reference
• Towards Efficient Data Valuation Based on the Shapley Value

  2019influential reference
• Variational Adversarial Active Learning

  2019influential reference
• BatchBALD: Efficient and Diverse Batch Acquisition for Deep Bayesian Active Learning

  2019cited by this paper
• Semi-Supervised Semantic Segmentation With High- and Low-Level Consistency

  2019cited by this paper
• Deep Active Learning: Unified and Principled Method for Query and Training

  2019cited by this paper
• Parting with Illusions about Deep Active Learning

  2019influential reference
• BDD100K: A Diverse Driving Video Database with Scalable Annotation Tooling

  2018cited by this paper
• Encoder-Decoder with Atrous Separable Convolution for Semantic Image Segmentation

  2018cited by this paper
• CEREALS - Cost-Effective REgion-based Active Learning for Semantic Segmentation

  2018cited by this paper
• Cost-Effective Active Learning for Deep Image Classification

  2017cited by this paper
• Deep Bayesian Active Learning with Image Data

  2017cited by this paper
• Active Learning for Convolutional Neural Networks: A Core-Set Approach

  2017influential reference
• Active and semi-supervised learning for object detection with imperfect data

  2017cited by this paper
• Wider or Deeper: Revisiting the ResNet Model for Visual Recognition

  2016cited by this paper
• DeepLab: Semantic Image Segmentation with Deep Convolutional Nets, Atrous Convolution, and Fully Connected CRFs

  2016cited by this paper
• The Cityscapes Dataset for Semantic Urban Scene Understanding

  2016cited by this paper
• Simple and Scalable Predictive Uncertainty Estimation using Deep Ensembles

  2016cited by this paper
• Active and Semi-Supervised Learning in ASR: Benefits on the Acoustic and Language Models

  2016cited by this paper
• Dropout as a Bayesian Approximation: Representing Model Uncertainty in Deep Learning

  2015cited by this paper
• Semi-Supervised Learning

  2014cited by this paper
• Active learning

  2013cited by this paper
• The MNIST Database of Handwritten Digit Images for Machine Learning Research [Best of the Web]

  2012cited by this paper
• Semantic contours from inverse detectors

  2011cited by this paper
• Bayesian Active Learning for Classification and Preference Learning

  2011cited by this paper
• The Pascal Visual Object Classes (VOC) Challenge

  2010cited by this paper
• Semantic object classes in video: A high-definition ground truth database

  2009cited by this paper
• ImageNet: A large-scale hierarchical image database

  2009cited by this paper
• A Mathematical Theory of Communication

  2006cited by this paper

• Active Learning with Task-Driven Representations for Messy Pools

  2025cites this paper
• CropLayer: a 2 m resolution cropland map of China for 2020 from Mapbox and Google satellite imagery

  2025cites this paper
• Exploring the effects of synthetic data generation: a case study on autonomous driving for semantic segmentation

  2025cites this paper
• Diffusion-Driven Two-Stage Active Learning for Low-Budget Semantic Segmentation

  2025cites this paper
• Active Learning Methods for Efficient Data Utilization and Model Performance Enhancement

  2025cites this paper
• Reducing Annotation Effort in Semantic Segmentation Through Conformal Risk Controlled Active Learning

  2025cites this paper
• Bayesian Active Learning for Semantic Segmentation

  2024cites this paper
• ARSFineTune: On-the-Fly Tuning of Vision Models for Unmanned Ground Vehicles

  2024cites this paper
• Utilizing active learning to accelerate segmentation of microstructures with tiny annotation budgets

  2024cites this paper
• Combining Image and Region Uncertainty-Based Active Learning for Melanoma Segmentation

  2024cites this paper
• Toward global rooftop PV detection with Deep Active Learning

  2024cites this paper
• ValUES: A Framework for Systematic Validation of Uncertainty Estimation in Semantic Segmentation

  2024cites this paper
• Key Patch Proposer: Key Patches Contain Rich Information

  2024cites this paper
• Making Better Use of Unlabelled Data in Bayesian Active Learning

  2024cites this paper
• TSynD: Targeted Synthetic Data Generation for Enhanced Medical Image Classification

  2024cites this paper
• Cropland inundation mapping in a mountain dominated region based on multi-resolution remotely sensed imagery and active learning for semantic segmentation

  2024cites this paper
• Hyperbolic Active Learning for Semantic Segmentation under Domain Shift

  2023cites this paper
• Synthetic Dataset Acquisition for a Specific Target Domain

  2023cites this paper
• Semi–Supervised Bayesian Active Learning with Task–Driven Representations

  year unknowncites this paper
• DSP ART : A L ARGE - SCALE D IFFUSION - GENERATED S YNTHETIC D ATASET WITH A NNOTATIONS FROM 3D P ARTS

  year unknowncites this paper
• CropLayer: A high-accuracy 2-meter resolution cropland mapping dataset for China in 2020 derived from Mapbox and Google satellite imagery using data-driven approaches

  year unknowncites this paper