百科.dev
全部条目在线工具AI 编程趋势榜开源项目技术资讯提交条目
登录
< 返回工具列表
L

LeYOLO

> 编程语言
开源

- 在 Google colab 中添加了 LeYOLO [](https://colab.research.google.com/GitHub/LilianHollard/LeYOLO/blob/main/minimal.ipynb) 抽象论文 (pr …

200 stars0 点赞1 次浏览
访问官网GitHub

工具介绍

- 在 Google colab 中添加了 LeYOLO [](https://colab.research.google.com/GitHub/LilianHollard/LeYOLO/blob/main/minimal.ipynb) 抽象论文 (pr …

  • Added LeYOLO in Google colab

Abstract

Paper (preprint!): https://arxiv.org/abs/2406.14239

Computational efficiency in deep neural networks is critical for object detection, especially as newer models prioritize speed over efficient computation (FLOP). This evolution has somewhat left behind embedded and mobile-oriented AI object detection applications. In this paper, we focus on design choices of neural network architectures for efficient object detection computation based on FLOP and propose several optimizations to enhance the efficiency of YOLO-based models.

Firstly, we introduce an efficient backbone scaling inspired by inverted bottlenecks and theoretical insights from the Information Bottleneck principle. Secondly, we present the Fast Pyramidal Architecture Network (FPAN), designed to facilitate fast multiscale feature sharing while reducing computational resources. Lastly, we propose a Decoupled Network-in-Network (DNiN) detection head engineered to deliver rapid yet lightweight computations for classification and regression tasks.

Building upon these optimizations and leveraging more efficient backbones, this paper contributes to a new scaling paradigm for object detection and YOLO-centric models called LeYOLO. Our contribution consistently outperforms existing models in various resource constraints, achieving unprecedented accuracy and flop ratio. Notably, LeYOLO-Small achieves a competitive mAP score of 38.2% on the COCO val with just 4.5 FLOP(G), representing a 42% reduction in computational load compared to the latest state-of-the-art YOLOv9-Tiny model while achieving similar accuracy. Our novel model family achieves a FLOP-to-accuracy ratio previously unattained, offering scalability that spans from ultra-low neural network configurations ( 4 GFLOPs) with 25.2, 31.3, 35.2, 38.2, 39.3 and 41 mAP for 0.66, 1.47, 2.53, 4.51, 5.8 and 8.4 FLOP(G).

LeYOLO Results

LeYOLO family model results on MSCOCO val dataset. Find pre-trained weights in the weights folder.

Models mAP Image Size FLOP (G)
LeYOLONano 25.2 320 0.66
LeYOLONano 31.3 480 1.47
LeYOLOSmall 35.2 480 2.53
LeYOLOSmall 38.2 640 4.51
LeYOLOMedium 39.3 640 5.80
LeYOLOLarge 41.0 768 8.40

Reproducibility

Reproduce LeYOLO results by using a cfg file. We did not use expansive or fancy training methods, just classic and default Ultralytics training recipes. We enable non-computer scientist to perform fast training reproducibility for their research with effortless training methods!

LeNeck

You can find the LeNeck combination with various backbones directly hard-coded within the Ultralytics environment. However, if you have your own backbone and training loop and want to understand/use the versatile object detectors, you can find the detailed code in the LeNeck/nn.py file. Using MobileNetv3 as an example, it explains how we join a backbone with LeYOLO.

LeYOLO installation and quickstart

Install

  • Warning ! Upgrading pip might be necessary
bash
python3 -m pip install --upgrade pip
  • Go into root folder
bash
pip install -e .

Quickstart

Use CLI or python interface to use LeYOLO / YOLOv8

  • Minimal example can be found in minimal.ipynb notebook.

Python

python
from ultralytics import YOLO

#Load LeYOLO model
model = YOLO("weights/LeYOLOSmall.pt")

#Use Ultralytics grounding computer vision methods
model.train(data="coco8.yaml", epochs=3)
model.val()
model.export(format="onnx")

Special thanks

We could only have accomplished this with the excellent work from the Ultralytics team since YOLOv5. LeYOLO fully integrates with Ultralytics, and their support and documentation comprehensively guide training, inference, validation, and export mechanisms.

For more details, refer to the Ultralytics documentation: https://www.ultralytics.com.

Contacts

Any questions or remarks? Contact [email protected]

Université de Reims Champagne-Ardenne - CEA, LRC DIGIT, LICIIS

UFR Sciences Exactes et Naturelles - Campus Moulin de la Housse BP 1039 - 51687 Reims Cedex 2 - Bat 2/3 - France https://www.univ-reims.fr

Issues· 2 开放

查看全部 Issues在 GitHub 打开

暂无开放 Issues,或尚未同步最近议题。

> 标签

Python

暂无评论,来聊聊你的看法吧

> 工具信息

发布日期2026年8月1日
最后更新2026年9月18日
分类编程语言
定价开源

> 相关工具

T
TypeScript
JavaScript 的超集,为前端与全栈提供静态类型
P
Python
通用编程语言,广泛用于 Web、数据与 AI
G
Go
Google 推出的简洁高效系统语言