Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

RoBiS: Robust Binary Segmentation for High-Resolution Industrial Images

Xurui Li, Zhonesheng Jiang, Tingxuan Ai, Yu Zhou

TL;DR

RoBiS tackles robustness gaps in unsupervised anomaly detection for real-world, high-resolution industrial imagery under varying lighting and viewpoint conditions. It fuses Swin-Cropping pre-processing, INP-Former-based anomaly detection with photometric augmentation, and a two-stage post-processing pipeline that combines adaptive binarization (MEBin) with SAM-Finer refinement to yield precise binary masks. The approach achieves state-of-the-art SegF1 on challenging MVTec AD 2 private test sets (e.g., 51.00% on TEST_priv and 46.52% on TEST_priv_mix), with strong cross-category robustness demonstrated across AucPro_0.05 and ClassF1. This work offers a practical, scalable solution for robust defect localization in diverse industrial categories and provides public code to facilitate adoption and benchmarking.

Abstract

Robust unsupervised anomaly detection (AD) in real-world scenarios is an important task. Current methods exhibit severe performance degradation on the MVTec AD 2 benchmark due to its complex real-world challenges. To solve this problem, we propose a robust framework RoBiS, which consists of three core modules: (1) Swin-Cropping, a high-resolution image pre-processing strategy to preserve the information of small anomalies through overlapping window cropping. (2) The data augmentation of noise addition and lighting simulation is carried out on the training data to improve the robustness of AD model. We use INP-Former as our baseline, which could generate better results on the various sub-images. (3) The traditional statistical-based binarization strategy (mean+3std) is combined with our previous work, MEBin (published in CVPR2025), for joint adaptive binarization. Then, SAM is further employed to refine the segmentation results. Compared with some methods reported by the MVTec AD 2, our RoBiS achieves a 29.2% SegF1 improvement (from 21.8% to 51.00%) on Test_private and 29.82% SegF1 gains (from 16.7% to 46.52%) on Test_private_mixed. Code is available at https://github.com/xrli-U/RoBiS.

RoBiS: Robust Binary Segmentation for High-Resolution Industrial Images

TL;DR

RoBiS tackles robustness gaps in unsupervised anomaly detection for real-world, high-resolution industrial imagery under varying lighting and viewpoint conditions. It fuses Swin-Cropping pre-processing, INP-Former-based anomaly detection with photometric augmentation, and a two-stage post-processing pipeline that combines adaptive binarization (MEBin) with SAM-Finer refinement to yield precise binary masks. The approach achieves state-of-the-art SegF1 on challenging MVTec AD 2 private test sets (e.g., 51.00% on TEST_priv and 46.52% on TEST_priv_mix), with strong cross-category robustness demonstrated across AucPro_0.05 and ClassF1. This work offers a practical, scalable solution for robust defect localization in diverse industrial categories and provides public code to facilitate adoption and benchmarking.

Abstract

Robust unsupervised anomaly detection (AD) in real-world scenarios is an important task. Current methods exhibit severe performance degradation on the MVTec AD 2 benchmark due to its complex real-world challenges. To solve this problem, we propose a robust framework RoBiS, which consists of three core modules: (1) Swin-Cropping, a high-resolution image pre-processing strategy to preserve the information of small anomalies through overlapping window cropping. (2) The data augmentation of noise addition and lighting simulation is carried out on the training data to improve the robustness of AD model. We use INP-Former as our baseline, which could generate better results on the various sub-images. (3) The traditional statistical-based binarization strategy (mean+3std) is combined with our previous work, MEBin (published in CVPR2025), for joint adaptive binarization. Then, SAM is further employed to refine the segmentation results. Compared with some methods reported by the MVTec AD 2, our RoBiS achieves a 29.2% SegF1 improvement (from 21.8% to 51.00%) on Test_private and 29.82% SegF1 gains (from 16.7% to 46.52%) on Test_private_mixed. Code is available at https://github.com/xrli-U/RoBiS.

Paper Structure

This paper contains 17 sections, 2 equations, 2 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Dataset Pre-processing
  4. Model Design
  5. Approach
  6. Architecture
  7. Training
  8. Results Post-processing
  9. Dataset & Evaluation
  10. Results
  11. Quantitative Results
  12. Qualitative Results
  13. Discussion
  14. Ablation study
  15. Future work
  16. ...and 2 more sections

Figures (2)

  • Figure 1: The pipeline of our RoBiS. This framework contains three important parts: (1) Swin-Cropping for dataset pre-processing (Sec. \ref{['data_preprocess']}). (2) INP-Former cvpr2025inp-former to detect anomalies (Sec. \ref{['detect_model']}). (3) Anomaly Maps Merging to obtain the anomaly map of original image, and Adaptive Binarization to generate its corresponding binary mask (Sec. \ref{['data_postprocess']}).
  • Figure 2: Visualization of anomaly segmentation results on the $\textit{TEST}_\text{pub}$ set of MVTec AD 2 dataset.