arXiv cs.AI· Johann Schmidt, Sebastian Stober, Joachim Denzler, Paul Bodesheim·· 15 小时前AI 评分22
PPS：面向细粒度分类的即插即用扫视视觉方案
PPS: Plug-and-Play Saccadic Vision for Fine-Grained Classification
AI 导读
研究者提出 PPS，一种即插即用的轻量扩展，可将任意带空间特征图的预训练骨干网络改造为扫视式分类器，仅新增一个优先级头和一个交叉注意力融合块。该方法用同一权重共享编码器处理降采样外周视图，由优先级头预测扫视优先级分布并并行编码高分辨率局部裁剪，复杂度随扫视次数线性增长，峰值内存与输入分辨率解耦。在九个基准上，该框架相对其骨干网络均取得一致提升。
正文 · 原文
Abstract:Fine-grained visual classification benefits from high-resolution inputs to resolve subtle, localized cues, but at the cost of large compute and memory budgets. Inspired by human saccadic vision, we propose a lightweight extension that turns any pretrained backbone with spatial feature maps into a saccadic classifier through glimpse-induced inference. The only architectural additions are a small priority head and a single cross-attention fusion block. The same weight-tied encoder encodes a downsampled peripheral glance, exposes the spatial feature map from which the priority head predicts a glimpse-priority distribution and re-encodes high-resolution glimpse crops sampled from that distribution in parallel. A sequential sampler with non-maximum suppression draws a runtime-configurable number of glimpses. A residual multi-head cross-attention block then fuses the glance against the glimpses. To improve the quality of the priority maps, we train the priority head by a per-glimpse information-gain target from a lightweight per-image memory bank. This requires no reinforcement learning. Complexity is linear in the number of glimpses and peak memory is decoupled from input resolution. Across nine benchmarks our framework improves consistently over its backbones.
| Subjects: | Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2509.15688 [cs.CV] |
| (or arXiv:2509.15688v2 [cs.CV] for this version) | |
| https://doi.org/10.48550/arXiv.2509.15688 arXiv-issued DOI via DataCite |
Submission history
From: Johann Schmidt johSchm [view email]
[v1]
Fri, 19 Sep 2025 07:03:37 UTC (11,385 KB)
[v2]
Thu, 29 Oct 2026 20:00:22 UTC (14,978 KB)
来源：arXiv cs.AI · arxiv.org