[论文解读] Efficient Teacher: Semi-Supervised Object Detection for YOLOv5
Efficient Teacher 引入 Dense Detector 基线,配备 Pseudo Label Assigner 和 Epoch Adaptor,以实现面向 YOLOv5 等一阶段锚点检测器的可扩展、高性能半监督目标检测。
Semi-Supervised Object Detection (SSOD) has been successful in improving the performance of both R-CNN series and anchor-free detectors. However, one-stage anchor-based detectors lack the structure to generate high-quality or flexible pseudo labels, leading to serious inconsistency problems in SSOD. In this paper, we propose the Efficient Teacher framework for scalable and effective one-stage anchor-based SSOD training, consisting of Dense Detector, Pseudo Label Assigner, and Epoch Adaptor. Dense Detector is a baseline model that extends RetinaNet with dense sampling techniques inspired by YOLOv5. The Efficient Teacher framework introduces a novel pseudo label assignment mechanism, named Pseudo Label Assigner, which makes more refined use of pseudo labels from Dense Detector. Epoch Adaptor is a method that enables a stable and efficient end-to-end semi-supervised training schedule for Dense Detector. The Pseudo Label Assigner prevents the occurrence of bias caused by a large number of low-quality pseudo labels that may interfere with the Dense Detector during the student-teacher mutual learning mechanism, and the Epoch Adaptor utilizes domain and distribution adaptation to allow Dense Detector to learn globally distributed consistent features, making the training independent of the proportion of labeled data. Our experiments show that the Efficient Teacher framework achieves state-of-the-art results on VOC, COCO-standard, and COCO-additional using fewer FLOPs than previous methods. To the best of our knowledge, this is the first attempt to apply Semi-Supervised Object Detection to YOLOv5.Code is available: https://github.com/AlibabaResearch/efficientteacher
研究动机与目标
- 激励面向一阶段锚点检测器的半监督目标检测(SSOD),并解决伪标签不一致性。
- 提出 Dense Detector,以密集输入将 YOLOv5 与 RetinaNet 的性能拉近。
- 引入 Pseudo Label Assigner,将伪标签分为可靠和不确定,以实现软监督引导。
- 开发 Epoch Adaptor,使端到端的 SSOD 训练在域适应和分布适应下高效进行。
- 展示在 VOC、COCO-standard、COCO-additional 上的最先进结果,且 FLOPs 更少。
提出的方法
- 通过在 RetinaNet 的基础上扩展,加入 YOLOv5 启发的密集采样和一个物体性分支,创建 Dense Detector。
- 开发 Pseudo Label Assigner,使用高/低阈值将伪标签分为可靠和不确定,并对不确定标签应用软损失。
- 将 SSOD 损失定义为 L = Ls + λLu,其中 Ls 在带标记数据上计算,Lu 在未标记数据上计算,使用类别、框和物体性项。
- 在 Lu 中,基于伪标签分数和对象性应用选择性损失,以缓解不可靠的伪标签。
- 引入 Epoch Adaptor,结合域自适应(通过梯度反转层)与分布自适应,按每个 epoch 动态设定伪标签阈值。
- 使用基于类别分布的分布自适应阈值 τ1^k、τ2^k,在各 epoch 平衡带标记/未标记数据。
实验结果
研究问题
- RQ1如何为一阶段锚点检测器在 SSOD 中细化伪标签分配?
- RQ2密集输入检测器(Dense Detector)是否能提升伪标签质量和 SSOD 性能,相较于传统的 RetinaNet 基线?
- RQ3分 epoch 的自适应策略是否能提升带未标记数据的 SSOD 训练稳定性与效率?
- RQ4相较于现有的 SSOD 方法,Efficient Teacher 在 VOC、COCO-standard、COCO-additional 上的增益是多少?
主要发现
- Efficient Teacher 在 VOC、COCO-standard、COCO-additional 上实现最先进的结果,且 FLOPs 少于先前方法。
- Dense Detector 相对于 RetinaNet 提升 AP50:95 5.36,同时推理 FLOPs 下降约 30%。
- Pseudo Label Assigner 通过将可靠和不确定的伪标签分开,并对后者应用软损失,显著改进 SSOD。
- Epoch Adaptor 具备域自适应和分布自适应,能加速训练、稳定学习,在更少的迭代中实现更好性能。
- 将 backbone 替换为 YOLOv5l 并在 Efficient Teacher 下训练,得到更好的最终检测性能,同时降低计算开销。
更好的研究,从现在开始
从论文设计到论文写作,大幅缩短您的研究时间。
无需绑定信用卡
本解读由 AI 生成,并经人工编辑审核。