Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

PP-YOLOE: An evolved version of YOLO

Shangliang Xu, Xinxin Wang, Wenyu Lv, Qinyao Chang, Cheng Cui, Kaipeng Deng, Guanzhong Wang, Qingqing Dang, Shengyu Wei, Yuning Du, Baohua Lai

TL;DR

This paper tackles real-time, high-accuracy object detection with deployment-friendly requirements by evolving PP-YOLOv2 into PP-YOLOE. It introduces an anchor-free detection framework built on a CSPRepResNet backbone/neck, augmented by TAL for dynamic label assignment and an Efficient Task-aligned Head (ET-head) to harmonize classification and localization. The approach yields notable performance gains, with PP-YOLOE-l achieving 51.4 mAP on COCO test-dev and 78.1 FPS on V100, and even higher speeds with TensorRT FP16, across a scalable family of models. The authors provide open-source PaddleDetection implementations and pretrained models, highlighting practical applicability across industrial settings.

Abstract

In this report, we present PP-YOLOE, an industrial state-of-the-art object detector with high performance and friendly deployment. We optimize on the basis of the previous PP-YOLOv2, using anchor-free paradigm, more powerful backbone and neck equipped with CSPRepResStage, ET-head and dynamic label assignment algorithm TAL. We provide s/m/l/x models for different practice scenarios. As a result, PP-YOLOE-l achieves 51.4 mAP on COCO test-dev and 78.1 FPS on Tesla V100, yielding a remarkable improvement of (+1.9 AP, +13.35% speed up) and (+1.3 AP, +24.96% speed up), compared to the previous state-of-the-art industrial models PP-YOLOv2 and YOLOX respectively. Further, PP-YOLOE inference speed achieves 149.2 FPS with TensorRT and FP16-precision. We also conduct extensive experiments to verify the effectiveness of our designs. Source code and pre-trained models are available at https://github.com/PaddlePaddle/PaddleDetection.

PP-YOLOE: An evolved version of YOLO

TL;DR

This paper tackles real-time, high-accuracy object detection with deployment-friendly requirements by evolving PP-YOLOv2 into PP-YOLOE. It introduces an anchor-free detection framework built on a CSPRepResNet backbone/neck, augmented by TAL for dynamic label assignment and an Efficient Task-aligned Head (ET-head) to harmonize classification and localization. The approach yields notable performance gains, with PP-YOLOE-l achieving 51.4 mAP on COCO test-dev and 78.1 FPS on V100, and even higher speeds with TensorRT FP16, across a scalable family of models. The authors provide open-source PaddleDetection implementations and pretrained models, highlighting practical applicability across industrial settings.

Abstract

In this report, we present PP-YOLOE, an industrial state-of-the-art object detector with high performance and friendly deployment. We optimize on the basis of the previous PP-YOLOv2, using anchor-free paradigm, more powerful backbone and neck equipped with CSPRepResStage, ET-head and dynamic label assignment algorithm TAL. We provide s/m/l/x models for different practice scenarios. As a result, PP-YOLOE-l achieves 51.4 mAP on COCO test-dev and 78.1 FPS on Tesla V100, yielding a remarkable improvement of (+1.9 AP, +13.35% speed up) and (+1.3 AP, +24.96% speed up), compared to the previous state-of-the-art industrial models PP-YOLOv2 and YOLOX respectively. Further, PP-YOLOE inference speed achieves 149.2 FPS with TensorRT and FP16-precision. We also conduct extensive experiments to verify the effectiveness of our designs. Source code and pre-trained models are available at https://github.com/PaddlePaddle/PaddleDetection.
Paper Structure (8 sections, 2 equations, 3 figures, 4 tables)

This paper contains 8 sections, 2 equations, 3 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Method
  3. A Brief Review of PP-YOLOv2
  4. Improvement of PP-YOLOE
  5. Experiment
  6. Implementation details
  7. Comparsion with Other SOTA Detectors
  8. Conclusion

Figures (3)

  • Figure 1: Comparison of the PP-YOLOE and other state-of-the-art models. PP-YOLOE-l achieves 51.4 mAP on COCO test-dev and 78.1 FPS on Tesla V100, obtains 1.9 AP and 9.2 FPS improvement compared with PP-YOLOv2huang2021pp-yolov2.
  • Figure 2: The model architecture of our PP-YOLOE. The backbone is CSPRepResNet, the neck is Path Aggregation Network (PAN), and the head is Efficient Task-aligned Head (ET-head).
  • Figure 3: Structure of our RepResBlock and CSPRepResStage