Aug 2026· Electronics· Vol 15, pp. 3447· 0 citations· 9 references
TL;DR
This work proposes CCM (Confidence-Calibrated Consistency Matching)—a per-class curriculum threshold, a dual strong-view consistency loss, and a smooth confidence weighting that softly admits borderline pseudo-labels—together with a unified view in which FixMatch, FlexMatch, SoftMatch, and CCM instantiate a single generalized weighting function.
Abstract
Labeling images is expensive, but unlabeled data is abundant. Semi-supervised learning (SSL) addresses this gap, though the dominant pseudo-labeling methods can suffer from confirmation bias—reinforcing their own confident-but-wrong predictions—most severely when labels are scarcest. Under a controlled, reproducible compute-constrained protocol on CIFAR-10, SVHN, and CIFAR-100, we examine which ingredients of consistency-based SSL actually help when as few as four labels per class are available. We propose CCM (Confidence-Calibrated Consistency Matching)—a per-class curriculum threshold, a dual strong-view consistency loss, and a smooth confidence weighting that softly admits borderline pseudo-labels—together with a unified view in which FixMatch, FlexMatch, SoftMatch, and CCM instantiate a single generalized weighting function. On CIFAR-10 with 40 labels, CCM reaches 35.14%, a significant improvement over the FlexMatch design it directly extends (+3.30 percentage points (pp), paired t-test p = 0.001). SoftMatch, re-trained under the identical budget, performs better still at the two smallest budgets (37.04% at 40 labels, p = 0.049), while CCM leads numerically at 4000 labels: the two smooth-weighting designs top the extreme-scarcity board—convergent evidence that the smoothness of the weighting function, more than the placement of its threshold, is the decisive design axis. We also report a negative result: cross-view agreement helps neither as an admission gate nor as reliability reweighting, reducing accuracy by up to 4.37 pp; agreement is a positive correctness signal, but its absolute level (approximately 44% correct among agreeing pseudo-labels) is too low to filter on safely. Curriculum thresholding, by contrast, hurts on the easier SVHN dataset and fails outright on CIFAR-100 when its per-class statistics become too thin. CCM adds no inference-time cost. We do not claim universality; we characterize when each ingredient helps within a single, identical-budget protocol.
Weakly supervised semantic segmentation (WSSS) with image-level labels is largely limited by the reliability of dense seed supervision. Existing CAM- and CLIP-based methods provide complementary localization cues, but their predictions are biased in different ways: classification-oriented cues are usually precise but incomplete, whereas alignment-oriented cues offer broader coverage but are more susceptible to contextual noise. In this letter, we propose Reliability-Calibrated Posterior Supervision (RCPS), a simple yet effective framework that uses discriminative classification evidence to calibrate broad CLIP-oriented cues. RCPS first constructs a semantic target through classification-regularized posterior calibration, avoiding direct commitment to either noisy alignment responses or incomplete classification activations. It then estimates pixel-wise reliability from cue self-certainty, classification–alignment consistency, and confidence-preserving disagreement, retaining potentially complementary evidence while suppressing uncertain responses. The resulting reliability-weighted objective learns dense seeds from calibrated soft targets. Experiments on PASCAL VOC 2012 and MS COCO 2014 show that RCPS consistently improves seed quality, pseudo-mask quality, and final segmentation performance over strong CAM- and CLIP-based baselines.
Xiao-Ya Sun, Xin Xu· IEEE Signal Processing Lette...· 0 citations
Semi-supervised adaptation of vision foundation models (VFMs) commonly freezes the pretrained backbone and updates lightweight modules such as LoRA. However, pseudo-labels have mixed reliability, and a single LoRA adapter must absorb reliable, ambiguous, and noisy gradients in the same low-rank space. This can make VFM adaptation sensitive to pseudo-label noise. We propose \textbf{TriNoL}, a \textbf{Tri}ple-expert learning framework from \textbf{No}isy \textbf{L}abels for semi-supervised VFM adaptation. TriNoL routes unlabeled samples into three confidence regions and assigns them to three LoRA experts: a Positive Expert for high-confidence pseudo-labels, an Alignment Expert for medium-confidence ambiguous samples, and a Negative Expert for low-confidence noisy samples. The VFM backbone remains frozen, and only the LoRA experts and classifier head are updated. By separating different pseudo-label reliability regions into specialized adaptation paths, TriNoL improves robustness to noisy supervision while keeping the training cost low.
Xuanyu Liu, Zheng Fang, Hongyang He et al.· 1 citation
The problem of Multi-Label Image Recognition with Partial Labels (MLIR-PL) is a significant challenge in computer vision, primarily due to the scarcity and high cost of complete annotations. Recent advances have leveraged large-scale vision-language models, such as CLIP, to establish rich correspondences between images and their labels, thereby improving the MLIR-PL performance. However, the existing CLIP-based methods have not fully exploited fine-grained local image features to mitigate interference from semantically irrelevant regions. Moreover, many studies have oversimplified the use of prompt contexts, limiting their ability to comprehensively capture the multi-dimensional attributes of categories. To address these limitations, this article proposes a novel MLIR-PL model with Visual–Label Alignment and Attribute-Aware Prompt (VA \({}^{3}\) P), which sufficiently harnesses the capabilities of large-scale pre-trained vision-language models. In the model, we design a Visual–Label Alignment module to establish a mapping between local image features and category text representations, conspicuously reducing the interference from irrelevant regions. Additionally, our Attribute-Aware Prompt module offers diverse contextual information, providing a more comprehensive representation of the category’s attributes. Extensive experimental results on the COCO 2014 and VOC 2007 datasets, compared with multiple state-of-the-art methods, demonstrate that our model achieves the best performance comprehensively, verifying the advantages of the proposed model in the MLIR-PL task.
Dengdi Sun, Hongxing Xie, Zhenbin Cai et al.· ACM Transactions on Multimed...· 0 citations
Zero-shot classification needs efficient label retrieval and fine-grained visual reasoning, yet discriminative and generative vision-language models fail in complementary ways.When CLIP's top-1 prediction is wrong, the correct label often remains in its top-$K$ shortlist, making disambiguation rather than recall the key challenge.Standalone generative models, however, are hindered by large label spaces and unconstrained outputs.This complementarity motivates separating broad candidate retrieval from fine-grained, image-grounded verification.We propose G2D, a training-free framework that uses a generative VLM to verify CLIP-retrieved candidates against the image.Candidate names and CLIP probabilities provide a structured prior for resolving visually similar classes.Fixed confidence routing, entropy-adaptive candidate sizing, and trie-constrained decoding focus generative reasoning on uncertain samples and ensure one valid output for each input at test time.Across eight benchmarks, G2D achieves 68.85% average accuracy, versus 59.35% for CLIP and 63.11% for the standalone VLM.Across seven generator configurations, candidate-set verification improves average accuracy by 1.08--27.42 percentage points.G2D also transfers to DCLIP, WaffleCLIP, and CuPL, supporting a practical interface between discriminative proposal and generative visual reasoning. Code: https://github.com/Harzva/G2D
Zehua Hao, Fang Liu, Qin-Liang Wang et al.· 0 citations
Most existing change detection methods rely on supervised learning and require costly pixel-level annotations. Weakly supervised change detection (WSCD) alleviates this burden by using only image-level labels. Existing image-level WSCD methods usually generate class activation maps (CAMs) through change classification and then obtain change maps by thresholding the CAMs. However, CAMs naturally highlight the most discriminative regions, which leads to a lack of sufficient spatial details. In addition, the appearance heterogeneity of land-cover objects easily causes class ambiguity in CAMs. To address these issues, we propose dual consistency learning (DUEL). DUEL introduces two consistency constraints, namely feature consistency and activation consistency, both guided by segment anything model (SAM)-derived object masks. Feature consistency uses mask-derived pairwise relation labels to perform semantic alignment in the bitemporal feature space, which helps alleviate class ambiguity. Activation consistency combines semantic cues from CAMs with spatial priors from SAM masks to generate pseudolabels for CAM regularization, with the goal of better recovering spatial details. Extensive experiments on two public change detection datasets demonstrate that DUEL outperforms several state-of-the-art WSCD methods.
Feng Zhou, Sun Zhang, Renlong Hang· IEEE Geoscience and Remote S...· 0 citations
Semi-supervised semantic segmentation has long turned on one question, which pseudo-labels to trust, and a generation of selection rules, dynamic thresholds, per-class curricula, soft confidence weights, answered it for the noisy, under-confident ResNet teachers of their day. Self-supervised foundation encoders change the regime: with a DINOv2 teacher, confidence saturates, so the filtering that helped a weak teacher can hurt a strong one. We propose CW-BASS v2, a saturation-aware pseudo-label selection method that reads the teacher's confidence regime rather than committing to one rule. It pairs held-out calibration, an unbiased per-class noise estimate, with a self-adaptive confidence floor that provably bounds retention away from 1, and combines them in a one-pass gate: measure the reliability of the teacher's confident set, pi_kept = Pr[correct | c>= tau], on a held-out slice, and filter strictly when it meets the confidence demanded (pi_kept>= tau), falling back to the adaptive floor otherwise. The boundary is the pre-existing operating threshold, not a value tuned to mIoU, and across six DINOv2 teachers it makes the correct strict-vs-floor call blind. CW-BASS v2 thus recovers the UniMatch V2 operating point on the saturated benchmarks by selecting strict (Pascal VOC 1/8 87.4 against its reported 87.9; Cityscapes within 0.5), and improves on it where the confident set is unreliable (pi_kept ~ 89%, ADE20K), where the floor edges ahead (+1.5 mIoU, single seed). The gate is principled because the failure it avoids is measured, not assumed: on a reliable, saturated teacher the confidence distribution's dynamic range collapses (98% of Pascal pixels>= 0.95), so an adaptive cutoff floods the retention mask and self-training decays into confirmation bias.
