arXiv cs.AI· Ebenezer Tarubinga·· 11 小时前AI 评分22
PixCon：面向基础模型半监督分割的干净正样本对比学习
PixCon: Clean-Positive Contrastive Learning for Foundation-Model Semi-Supervised Segmentation
AI 导读
PixCon 在 DINOv2-Base 版 UniMatch V2 上加入类级记忆库，仅存储学生预测与标签一致的标注像素特征，用多正样本对比损失训练共享编码器-解码器，推理时丢弃头和记忆库。
正文 · 原文
Abstract:Can a simple labeled-pixel contrastive branch improve semi-supervised segmentation with a strong pretrained encoder? We study this question with PixCon, which adds a class-wise memory bank to a DINOv2-Base implementation of UniMatch V2. The bank stores ground-truth-labeled features only where the student prediction agrees with the label. A multi-positive contrastive loss uses these features to train the shared encoder-decoder; the head and bank are discarded at inference. This simplifies earlier labeled-pixel banks by removing confidence thresholds and learned ranking. On Pascal VOC with 183 labels, PixCon improves all three paired seeds by 0.20, 0.64 and 1.83 mIoU points, raising the mean from 87.01 to 87.90. With 92 labels the mean gain is 0.40 points; single-seed differences on Cityscapes and ADE20K range from -0.08 to +0.14 points. A predicted-label contrastive control reaches nearly the same Pascal accuracy, so these results support the added branch rather than an accuracy benefit from its correctness filter. An exact gradient decomposition distinguishes mislabeled-positive counts from their similarity-weighted influence. Separate teacher diagnostics find retained-pixel error rates of 1.8% on Pascal and 10.6% on ADE20K. These diagnostics do not measure the training bank or predict final accuracy. PixCon adds 15.3-19.7% iteration time in the measured runs and no inference computation. We report paired seeds, training cost and the limits of the evidence.
| Comments: | 17 pages, 7 figures, 7 tables; main paper with supplementary material appended |
| Subjects: | Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Machine Learning (cs.LG); Image and Video Processing (eess.IV) |
| MSC classes: | 68T45, 68T07 |
| ACM classes: | I.4.6; I.2.10 |
| Cite as: | arXiv:2607.03068 [cs.CV] |
| (or arXiv:2607.03068v2 [cs.CV] for this version) | |
| https://doi.org/10.48550/arXiv.2607.03068 arXiv-issued DOI via DataCite |
Submission history
From: Ebenezer Tarubinga [view email]
[v1]
Fri, 3 Jul 2026 08:01:37 UTC (16,882 KB)
[v2]
Wed, 4 Nov 2026 05:32:51 UTC (13,539 KB)
来源：arXiv cs.AI · arxiv.org