Training-free Boost for Open-Vocabulary Object Detection with Confidence Aggregation

📄 arXiv: 2404.08603v1 📥 PDF

作者: Yanhao Zheng, Kai Liu

分类: cs.CV

发布日期: 2024-04-12

💡 一句话要点

提出无训练增益方法以解决开放词汇目标检测中的置信度聚合问题

🎯 匹配领域: 支柱三：空间感知与语义 (Perception & Semantics)

关键词: 开放词汇检测 目标检测 置信度聚合 无训练方法 视觉相似度 区域提议 文本引导

📋 核心要点

  1. 现有的开放词汇目标检测方法在处理新类别时，通常会因训练阶段的偏差而导致较低的置信度分数，影响检测效果。
  2. 本文提出了一种无训练的后处理方案AggDet，通过调整置信度分数和保留被错误丢弃的对象，提升新类别的检测性能。
  3. 在OV-COCO和OV-LVIS基准测试中，AggDet分别提升了3.3%和1.5%的检测精度，显示出其有效性。

📝 摘要（中文）

开放词汇目标检测（OVOD）旨在定位和识别训练时未见的新类别视觉对象。然而，实证研究表明，先进的检测器通常对这些新实例分配较低的分数，导致在推理过程中被常用的贪婪策略（如非极大值抑制NMS）抑制，从而影响新类别的检测性能。本文系统研究了这一问题，并提出了两种先进的措施来调整置信度分数并保留被错误丢弃的对象：通过区域/对象提议的重叠度估计类无关的定位质量，以及使用代理原型的文本引导视觉相似度估计。结合这些技术，本文提出了一种通用的、无训练的后处理方案AggDet，显著提升了开放词汇检测器的性能。

🔬 方法详解

问题定义：本文解决的是开放词汇目标检测中，训练阶段对新类别的低置信度评分问题，导致在推理时被错误抑制。

核心思路：通过引入类无关的定位质量估计和文本引导的视觉相似度估计，调整置信度分数，从而保留被错误丢弃的对象。

技术框架：整体框架包括区域提议阶段和对象分类阶段，分别针对这两个阶段设计了调整技术，以生成聚合的置信度估计。

关键创新：最重要的创新在于提出了AggDet这一无训练的后处理方案，能够在不同模型规模和架构设计中一致提升开放词汇检测器的性能。

关键设计：在区域提议阶段，使用重叠度来估计定位质量；在对象分类阶段，利用代理原型进行文本引导的视觉相似度估计，确保新类别的有效识别。

🖼️ 关键图片

fig_0
fig_1
fig_2

📊 实验亮点

在OV-COCO和OV-LVIS基准测试中，AggDet分别实现了3.3%和1.5%的性能提升，显示出其在开放词汇目标检测中的显著效果，且无需额外的训练成本。

🎯 应用场景

该研究的潜在应用领域包括智能监控、自动驾驶、机器人视觉等，能够有效提升系统对新类别物体的检测能力，具有重要的实际价值和未来影响。通过无训练的后处理方案，能够降低模型部署的复杂性，提升实时检测的准确性。

📄 摘要（原文）

Open-vocabulary object detection (OVOD) aims at localizing and recognizing visual objects from novel classes unseen at the training time. Whereas, empirical studies reveal that advanced detectors generally assign lower scores to those novel instances, which are inadvertently suppressed during inference by commonly adopted greedy strategies like Non-Maximum Suppression (NMS), leading to sub-optimal detection performance for novel classes. This paper systematically investigates this problem with the commonly-adopted two-stage OVOD paradigm. Specifically, in the region-proposal stage, proposals that contain novel instances showcase lower objectness scores, since they are treated as background proposals during the training phase. Meanwhile, in the object-classification stage, novel objects share lower region-text similarities (i.e., classification scores) due to the biased visual-language alignment by seen training samples. To alleviate this problem, this paper introduces two advanced measures to adjust confidence scores and conserve erroneously dismissed objects: (1) a class-agnostic localization quality estimate via overlap degree of region/object proposals, and (2) a text-guided visual similarity estimate with proxy prototypes for novel classes. Integrated with adjusting techniques specifically designed for the region-proposal and object-classification stages, this paper derives the aggregated confidence estimate for the open-vocabulary object detection paradigm (AggDet). Our AggDet is a generic and training-free post-processing scheme, which consistently bolsters open-vocabulary detectors across model scales and architecture designs. For instance, AggDet receives 3.3% and 1.5% gains on OV-COCO and OV-LVIS benchmarks respectively, without any training cost.