TriLite：基于通用视觉特征与三区域解耦的高效弱监督目标定位
TriLite: Efficient Weakly Supervised Object Localization with Universal Visual Features and Tri-Region Disentanglement
用冻结的 DINOv2 ViT 特征配合极少量可训练参数（<800K）做单阶段弱监督定位，TriHead 把图像块特征解耦为前景/背景/模糊三区以提升物体覆盖度，为「冻结大骨干+轻量头」的定位与密集预测任务提供了可复用范式。
原文依据 leverages a frozen Vision Transformer with DINOv2 pre-training in a self-supervised manner
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论弱监督目标定位（WSOL）旨在仅使用图像级标签在图像中定位目标物体。尽管近期取得了进展，许多方法仍依赖多阶段流程或对大型骨干网络进行全量微调，增加了训练成本，同时更广泛的 WSOL 社区仍面临物体覆盖不完整的挑战。我们提出 TriLite，一个单阶段 WSOL 框架，以自监督方式利用经 DINOv2 预训练的冻结视觉 Transformer，并仅引入极少量可训练参数（在 ImageNet-1K 上少于 800K）用于分类与定位。其核心是所提出的 TriHead 模块，它将图像块特征分解为前景、背景和模糊区域，从而在抑制虚假激活的同时提升物体覆盖度。通过解耦分类与定位目标，TriLite 有效利用了自监督 ViT 学到的通用表征，而无需昂贵的端到端训练。在 CUB-200-2011、ImageNet-1K 和 OpenImages 上的大量实验表明，TriLite 刷新了当前最优性能，同时相比先前方法显著更具参数效率且更易训练。代码见：https://github.com/ariansabaghi/TriLite
英文原文摘要 1317 字符
Weakly supervised object localization (WSOL) aims to localize target objects in images using only image-level labels. Despite recent progress, many approaches still rely on multi-stage pipelines or full fine-tuning of large backbones, increasing training cost, while the broader WSOL community continues to face the challenge of partial object coverage. We present TriLite, a single-stage WSOL framework that leverages a frozen Vision Transformer with DINOv2 pre-training in a self-supervised manner, and introduces only a minimal number of trainable parameters (fewer than 800K on ImageNet-1K) for both classification and localization. At its core is the proposed TriHead module, which decomposes patch features into foreground, background, and ambiguous regions, thereby improving object coverage while suppressing spurious activations. By disentangling classification and localization objectives, TriLite effectively exploits the universal representations learned by self-supervised ViTs without requiring expensive end-to-end training. Extensive experiments on CUB-200-2011, ImageNet-1K, and OpenImages demonstrate that TriLite sets a new state of the art, while remaining significantly more parameter-efficient and easier to train than prior methods. Code is available at: https://github.com/ariansabaghi/TriLite