面向全景街景分割的 Vision Transformer 集成方法（PalmCity 挑战赛）
Vision Transformer Ensembles for Panoramic Street Segmentation
AI 导读
一篇提交 PalmCity 挑战赛的研究记录了两阶段全景街景语义分割方案：第一阶段比较九种预训练分割系统，三个采用 DINOv3 特征的 Encoder only Mask Transformer 以三尺度加水平翻转评估，在 84 张图验证集取得 60.95% mIoU，并曾在 2026 年 10 月 5 日快照的隐藏测试榜居首。
正文 · 原文
Abstract:Semantic segmentation of street panoramas can support detailed descriptions of urban environments, yet small datasets and unequal training costs make model selection difficult. This paper documents the two stages of a submission to the PalmCity challenge. In the first stage, nine pretrained segmentation systems are compared using approximately equal computation budgets, and the two leading candidates are trained with three random seeds. An equal probability average of three Encoder only Mask Transformer models with DINOv3 features, evaluated at three image scales with horizontal reflection, reaches 60.95% mean intersection over union on the 84 image validation split and 57.08% on the hidden test leaderboard, first in the snapshot of 5 October 2026. In the second stage, color and crop augmentation, rare class sampling, larger inputs, further architectures, and post processing are tested under the same update budget, and candidates are compared through a paired image bootstrap with selection rules fixed before the results. Training at 768 by 1536 and 896 by 1792 pixels gives the largest gains. The selected ensemble of four models scores 64.53% on validation when trained on the training images alone. Retrained on the training and validation images, it receives 61.54% mean intersection over union and 71.93% mean F1 on the hidden test leaderboard, first among five entries in the snapshot of 30 October 2026. Segment Anything voting, instance mask overlays, and ensembles with other architectures all reduce the validation score. Because the validation images guided many choices and the final models were trained on them, the validation scores are optimistic and the test score is the only independent measurement. The study reports all runs, comparisons, costs, and limits, and releases the code and models.
| Comments: | 26 pages, 6 figures, 12 tables. Version 2 adds the second stage of 28 to 30 October 2026 with its hidden test result, negative results and limits |
| Subjects: | Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Machine Learning (cs.LG) |
| Cite as: | arXiv:2610.06063 [cs.CV] |
| (or arXiv:2610.06063v2 [cs.CV] for this version) | |
| https://doi.org/10.48550/arXiv.2610.06063 arXiv-issued DOI via DataCite |
Submission history
From: Yunus Serhat Bıçakçı [view email]
[v1]
Mon, 5 Oct 2026 09:56:48 UTC (6,143 KB)
[v2]
Fri, 30 Oct 2026 13:13:43 UTC (7,058 KB)
来源：arXiv cs.AI · arxiv.org