Accuracy Improvement of Semi-Supervised Segmentation Using Supervised ClassMix and Sup-Unsup Feature Discriminator

Takahiro Mano; Reiji Saito; Kazuhiro Hotta

cs.CVApr 8, 2026

Accuracy Improvement of Semi-Supervised Segmentation Using Supervised ClassMix and Sup-Unsup Feature Discriminator

Authors: Takahiro Mano, Reiji Saito, Kazuhiro Hotta

Organizations: Meijo University, 1-501 Shiogamaguchi, Tempaku-ku, Nagoya 468-8502, Japan

Abstract

In semantic segmentation, the creation of pixel-level labels for training data incurs significant costs. To address this problem, semi-supervised learning, which utilizes a small number of labeled images alongside unlabeled images to enhance the performance, has gained attention. A conventional semi-supervised learning method, ClassMix, pastes class labels predicted from unlabeled images onto other images. However, since ClassMix performs operations using pseudo-labels obtained from unlabeled images, there is a risk of handling inaccurate labels. Additionally, there is a gap in data quality between labeled and unlabeled images, which can impact the feature maps. This study addresses these two issues. First, we propose a method where class labels from labeled images, along with the corresponding image regions, are pasted onto unlabeled images and their pseudo-labeled images. Second, we introduce a method that trains the model to make predictions on unlabeled images more similar to those on labeled images. Experiments on the Chase and COVID-19 datasets demonstrated an average improvement of 2.07% in mIoU compared to conventional semi-supervised learning methods.

Figures & tables

Figure 1: Comparison between the conventional ClassMix and the proposed SupMix. ClassMix is mixing images between images and their segmentation predictions. In contrast, SupMix pastes regions from a few classes in labeled images onto the pseudo-labels of unlabeled images. The accuracy of the pseudo-labels mixed by SupMix is improved by using labeled images and pasting them onto unlabeled images with pseudo-labels. Since the ground-truth labels are independent of the prediction accuracy, only a few classes must be pasted onto the unlabeled images and pseudo-labels.

Figure 2: The Sup-Unsup Feature Discriminator involves a shared encoder-decoder architecture used for both supervised learning and pseudo-supervised learning. Note that $x_{l}$ represents the labeled image, $x_{w}$ represents the unlabeled image (weak augmentation), and $x_{s}$ represents the unlabeled image (strong augmentation). $p_{l}$ denotes the output feature map of the labeled image, $p_{w}$ denotes the output feature map of the unlabeled image (weak augmentation), and $p_{s}$ denotes the output feature map of the unlabeled image (strong augmentation). Additionally, y represents the ground truth. During training, a discriminator is employed to make it difficult to distinguish between the supervised features and the unsupervised features without fixing, which are obtained after applying weak augmentations.

Chase	1/8 (2 images)				1/4 (5 images)
	Supervised	FixMatch	UniMatch	ours	Supervised	FixMatch	UniMatch	ours
background	74.62 $\pm$ 37.31	95.6 $\pm$ 0.12	96.41 $\pm$ 0.06	96.58 $\pm$ 0.06	39.12 $\pm$ 44.57	96.48 $\pm$ 0.05	96.50 $\pm$ 0.02	96.63 $\pm$ 0.03
retinal vessel	1.86 $\pm$ 2.54	38.54 $\pm$ 2.86	55.61 $\pm$ 0.81	58.91 $\pm$ 0.31	3.78 $\pm$ 3.09	57.62 $\pm$ 0.60	57.78 $\pm$ 0.35	60.41 $\pm$ 0.20
mean IoU	38.24 $\pm$ 17.51	67.07 $\pm$ 1.49	76.01 $\pm$ 0.43	77.74 $\pm$ 0.18	21.45 $\pm$ 23.83	77.05 $\pm$ 0.32	77.14 $\pm$ 0.18	78.52 $\pm$ 0.10
COVID-19	1/8 (9 images)				1/4 (18 images)
background	96.26 $\pm$ 0.07	95.57 $\pm$ 0.23	96.65 $\pm$ 0.13	96.72 $\pm$ 0.11	96.78 $\pm$ 0.12	96.65 $\pm$ 0.08	97.03 $\pm$ 0.05	97.08 $\pm$ 0.07

Table 1: Accuracy on Chase and COVID-19 datasets for Supervised, FixMatch, UniMatch, and ours. The top table shows results on Chase and the bottom table shows the results on COVID-19 dataset. Each row shows the IoU and standard deviation for each class, while each column compares the case where the labeled data covers 1/8 (N images) and 1/4 (N images) in the total dataset.

Figure 3: Segmentation results on Chase and COVID-19 datasets. Each row shows the dataset name and the ratio of supervised images used for training in all supervised images. Each column shows Input Images (Input), Ground Truth (GT), and the results of Supervised, FixMatch, UniMatch, and Ours. In the Chase dataset, the background class is visualized in black, and the retinal vessel class is in white. In the COVID-19 dataset, black represents the background class, blue indicates the ground-glass class, green shows the consolidation class, and red denotes the pleural effusions class.

Methods	CutMix	ClassMix	SupMix	SUFD	ours	CutMix	ClassMix	SupMix	SUFD	ours
Chase	1/8 (2 images)					1/4 (5 images)
retinal vessel	55.61 ( $+$ 0.00)	56.78 ( $+$ 1.17)	57.57 ( $+$ 1.96)	55.95 ( $+$ 0.34)	58.91 ( $+$ 3.30)	57.78 ( $+$ 0.00)	57.91 ( $+$ 0.13)	59.60 ( $+$ 1.82)	56.90 ( $-$ 0.88)	60.41 ( $+$ 2.63)
COVID-19	1/8 (9 images)					1/4 (18 images)
ground-glass	26.55 ( $+$ 0.00)	27.95 ( $+$ 1.40)	32.82 ( $+$ 6.27)	33.36 ( $+$ 6.81)	37.25 ( $+$ 10.70)	34.09 ( $+$ 0.00)	34.99 ( $+$ 0.90)	38.10 ( $+$ 4.01)	34.17 ( $+$ 0.07)	38.85 ( $+$ 4.76)

Table 2: Verification of the individual effects of SupMix and SUFD. Each row indicates the dataset type and the classes with fewer samples (retinal vessel and ground-glass). In each column, the results display the accuracy when we train the network with 1/N images in all supervised images. The values in parentheses are the improvement over CutMix. The comparison methods include CutMix, ClassMix, SupMix, SUFD, and ours. CutMix corresponds to the standard UniMatch results, while “ours” refers to the combined method of SupMix and SUFD.

Semi-Supervised Medical Image Segmentation

Semi-supervised medical image segmentation uses labeled and unlabeled scans to train models that delineate organs, lesions, and tissues with fewer annotations.

See all 49 papers

SSS: Semi-Supervised SAM-2 with Efficient Prompting for Medical Imaging Segmentation
Jun 10, 2025Hongjie Zhu, Xiwei Liu, Rundong Xue +5Promptable Image Segmentation Semi-Supervised Medical Image Segmentation
SWUN · MBZUAI · XJTU +2
SemiGDA: Generative Dual-distribution Alignment for Semi-Supervised Medical Image Segmentation
Apr 25, 2026Kaiwen Huang, Yi Zhou, Yizhe Zhang +2Distribution Matching Semi-Supervised Medical Image Segmentation
Nanjing University of Science and Technology, Nanjing, China · Southeast University, Nanjing, China
HDVS: semi-supervised semantic segmentation via heterogeneous dual-branch voting supervision
Oct 27, 2026Yongqi Shan, Yunzhi Zhuge, Huchuan LuSemantic Segmentation Pseudo-Labeling
School of Computer Science and Technology, Dalian University of Technology, Dalian 116081, China. · School of Information and Communication Engineering, Dalian University of Technology, Dalian 116081, China.

Jun 10, 2025J/K move · Enter open · S save

SSS: Semi-Supervised SAM-2 with Efficient Prompting for Medical Imaging Segmentation

Hongjie Zhu, Xiwei Liu, Rundong Xue, Zeyu Zhang+4

SWUN · MBZUAI · XJTU+2

Promptable Image Segmentation Semi-Supervised Medical Image Segmentation Medical Image Segmentation Semi-Supervised Learning Segment Anything Models

Open paper AskPDF arXiv

In the era of information explosion, efficiently leveraging large-scale unlabeled data while minimizing the reliance on high-quality pixel-level annotations remains a critical challenge in the field of medical imaging. Semi-supervised learning (SSL) enhances the utilization of unlabeled data by facilitating knowledge transfer, significantly improving the performance of fully supervised models and emerging as a highly promising research direction in medical image analysis. Inspired by the ability of Vision Foundation Models (e.g., SAM-2) to provide rich prior knowledge, we propose SSS (Semi-Supervised SAM-2), a novel approach that leverages SAM-2's robust feature extraction capabilities to uncover latent knowledge in unlabeled medical images, thus effectively enhancing feature support for fully supervised medical image segmentation. Specifically, building upon the single-stream "weak-to-strong" consistency regularization framework, this paper introduces a Discriminative Feature Enhancement (DFE) mechanism to further explore the feature discrepancies introduced by various data augmentation strategies across multiple views. By leveraging feature similarity and dissimilarity across multi-scale augmentation techniques, the method reconstructs and models the features, thereby effectively optimizing the salient regions. Furthermore, a prompt generator is developed that integrates Physical Constraints with a Sliding Window (PCSW) mechanism to generate input prompts for unlabeled data, fulfilling SAM-2's requirement for additional prompts. Extensive experiments demonstrate the superiority of the proposed method for semi-supervised medical image segmentation on two multi-label datasets, i.e., ACDC and BHSD. Notably, SSS achieves an average Dice score of 53.15 on BHSD, surpassing the previous state-of-the-art method by +3.65 Dice. Code will be available at https://github.com/AIGeeksGroup/SSS.

Abstract

Semi-Supervised Medical Image Segmentation

Explore similar work

SSS: Semi-Supervised SAM-2 with Efficient Prompting for Medical Imaging Segmentation

SemiGDA: Generative Dual-distribution Alignment for Semi-Supervised Medical Image Segmentation

HDVS: semi-supervised semantic segmentation via heterogeneous dual-branch voting supervision