https://github.com/ultralytics/yolov3

Introduction简介

This directory contains PyTorch YOLOv3 software developed by Ultralytics LLC, and is freely available for redistribution under the GPL-3.0 license. For more information please visit https://www.ultralytics.com.

此目录包含由Ultralytics LLC开发的PyTorch YOLOv3软件，可根据GPL-3.0许可证免费重新分发。更多信息请访问https://www.ultralytics.com。

Description描述

The https://github.com/ultralytics/yolov3 repo contains inference and training code for YOLOv3 in PyTorch. The code works on Linux, MacOS and Windows. Training is done on the COCO dataset by default: https://cocodataset.org/#home. Credit to Joseph Redmon for YOLO: https://pjreddie.com/darknet/yolo/.

https://github.com/ultralytics/yolov3 托管包含PyTorch中yolov3的引用和训练代码。这段代码可以在Linux、MacOS和Windows上运行。默认情况下，训练是在COCO数据集上完成的：https://COCO dataset.org/'35；home。YOLO的版权归属于约瑟夫·雷蒙：https://pjreddie.com/darknet/YOLO/。

Requirements

Python 3.7 or later with the following pip3 install -U -r requirements.txt packages:

  • numpy
  • torch >= 1.1.0
  • opencv-python
  • tqdm

Tutorials教程

Jupyter Notebook

Our Jupyter notebook provides quick training, inference and testing examples.

我们的Jupyter笔记本提供了快速的培训、推理和测试示例。

Training

Start Training: python3 train.py to begin training after downloading COCO data with data/get_coco_dataset.sh. Each epoch trains on 117,263 images from the train and validate COCO sets, and tests on 5000 images from the COCO validate set.

在使用data/get_coco_dataset.sh下載了COCO數據集之後使用python3 train.py開始訓練。

Resume Training: python3 train.py --resume to resume training from weights/last.pt.

Plot Training: from utils import utils; utils.plot_results() plots training results from coco_16img.data, coco_64img.data, 2 example datasets available in the data/ folder, which train and test on the first 16 and 64 images of the COCO2014-trainval dataset.

Image Augmentation

datasets.py applies random OpenCV-powered (https://opencv.org/) augmentation to the input images in accordance with the following specifications. Augmentation is applied only during training, not during inference. Bounding boxes are automatically tracked and updated with the images. 416 x 416 examples pictured below.

Augmentation Description
Translation +/- 10% (vertical and horizontal)
Rotation +/- 5 degrees
Shear +/- 2 degrees (vertical and horizontal)
Scale +/- 10%
Reflection 50% probability (horizontal-only)
HSV Saturation +/- 50%
HSV Intensity +/- 50%

Speed

https://cloud.google.com/deep-learning-vm/
Machine type: preemptible n1-standard-16 (16 vCPUs, 60 GB memory)
CPU platform: Intel Skylake
GPUs: K80 ($0.20/hr), T4 ($0.35/hr), V100 ($0.83/hr) CUDA with Nvidia Apex FP16/32
HDD: 1 TB SSD
Dataset: COCO train 2014 (117,263 images)
Model: yolov3-spp.cfg
Command: python3 train.py --img 416 --batch 32 --accum 2

GPU n --batch --accum img/s epoch
time		 epoch
cost
K80 1 32 x 2 11 175 min $0.58
T4 1
2		 32 x 2
64 x 1		 41
61		 48 min
32 min		 $0.28
$0.36
V100 1
2		 32 x 2
64 x 1		 122
178		 16 min
11 min		 $0.23
$0.31
2080Ti 1
2		 32 x 2
64 x 1		 81
140		 24 min
14 min		 -

Inference

detect.py runs inference on any sources:

  1. python3 detect.py --source ...
  • Image: --source file.jpg
  • Video: --source file.mp4
  • Directory: --source dir/
  • Webcam: --source 0
  • RTSP stream: --source rtsp://170.93.143.139/rtplive/470011e600ef003a004ee33696235daa
  • HTTP stream: --source http://wmccpinetop.axiscam.net/mjpg/video.mjpg

To run a specific models:

YOLOv3: python3 detect.py --cfg cfg/yolov3.cfg --weights yolov3.weights

YOLOv3-tiny: python3 detect.py --cfg cfg/yolov3-tiny.cfg --weights yolov3-tiny.weights

YOLOv3-SPP: python3 detect.py --cfg cfg/yolov3-spp.cfg --weights yolov3-spp.weights

Pretrained Weights

Download from: https://drive.google.com/open?id=1LezFG5g3BCW6iYaV89B2i64cqEUZD7e0

Darknet Conversion

  1. $ git clone https://github.com/ultralytics/yolov3 && cd yolov3
  2.  
  3. # convert darknet cfg/weights to pytorch model
  4. $ python3  -c "from models import *; convert('cfg/yolov3-spp.cfg', 'weights/yolov3-spp.weights')"
  5. Success: converted 'weights/yolov3-spp.weights' to 'converted.pt'
  6.  
  7. # convert cfg/pytorch model to darknet weights
  8. $ python3  -c "from models import *; convert('cfg/yolov3-spp.cfg', 'weights/yolov3-spp.pt')"
  9. Success: converted 'weights/yolov3-spp.pt' to 'converted.weights'

mAP

  1. python3 test.py --weights ... --cfg ...
  • mAP@0.5 run at --nms-thres 0.5, mAP@0.5...0.95 run at --nms-thres 0.7
  • YOLOv3-SPP ultralytics is ultralytics68.pt with yolov3-spp.cfg
  • Darknet results: https://arxiv.org/abs/1804.02767
  Size COCO mAP
@0.5...0.95		 COCO mAP
@0.5
YOLOv3-tiny
YOLOv3
YOLOv3-SPP
YOLOv3-SPP ultralytics		 320 14.0
28.7
30.5
35.4		 29.1
51.8
52.3
54.3
YOLOv3-tiny
YOLOv3
YOLOv3-SPP
YOLOv3-SPP ultralytics		 416 16.0
31.2
33.9
39.0		 33.0
55.4
56.9
59.2
YOLOv3-tiny
YOLOv3
YOLOv3-SPP
YOLOv3-SPP ultralytics		 512 16.6
32.7
35.6
40.3		 34.9
57.7
59.5
60.6
YOLOv3-tiny
YOLOv3
YOLOv3-SPP
YOLOv3-SPP ultralytics		 608 16.6
33.1
37.0
40.9		 35.4
58.2
60.7
60.9		 
  1. $ python3 test.py --save-json --img-size 608 --nms-thres 0.5 --weights ultralytics68.pt
  2.  
  3. Namespace(batch_size=16, cfg='cfg/yolov3-spp.cfg', conf_thres=0.001, data='data/coco.data', device='1', img_size=608, iou_thres=0.5, nms_thres=0.7, save_json=True, weights='ultralytics68.pt')
  4. Using CUDA device0 _CudaDeviceProperties(name='GeForce RTX 2080 Ti', total_memory=11019MB)
  5.  
  6.                Class    Images   Targets         P         R   mAP@0.5        F1: 100%|███████████████████████████████████████████████████████████████████████████████████| 313/313 [09:46<00:00,  1.09it/s]
  7.                  all     5e+03  3.58e+04    0.0823     0.798     0.595     0.145
  8.               person     5e+03  1.09e+04    0.0999     0.903     0.771      0.18
  9.              bicycle     5e+03       316    0.0491     0.782      0.56    0.0925
  10.                  car     5e+03  1.67e+03    0.0552     0.845     0.646     0.104
  11.           motorcycle     5e+03       391      0.11     0.847     0.704     0.194
  12.             airplane     5e+03       131     0.099     0.947     0.878     0.179
  13.                  bus     5e+03       261     0.142     0.874     0.825     0.244
  14.                train     5e+03       212     0.152     0.863     0.806     0.258
  15.                truck     5e+03       352    0.0849     0.682     0.514     0.151
  16.                 boat     5e+03       475    0.0498     0.787     0.504    0.0937
  17.        traffic light     5e+03       516    0.0304     0.752     0.516    0.0584
  18.         fire hydrant     5e+03        83     0.144     0.916     0.882     0.248
  19.            stop sign     5e+03        84    0.0833     0.917     0.809     0.153
  20.        parking meter     5e+03        59    0.0607     0.695     0.611     0.112
  21.                bench     5e+03       473    0.0294     0.685     0.363    0.0564
  22.                 bird     5e+03       469    0.0521     0.716     0.524    0.0972
  23.                  cat     5e+03       195     0.252     0.908      0.78     0.395
  24.                  dog     5e+03       223     0.192     0.883     0.829     0.315
  25.                horse     5e+03       305     0.121     0.911     0.843     0.214
  26.                sheep     5e+03       321     0.114     0.854     0.724     0.201
  27.                  cow     5e+03       384     0.105     0.849     0.695     0.187
  28.             elephant     5e+03       284     0.184     0.944     0.912     0.308
  29.                 bear     5e+03        53     0.358     0.925     0.875     0.516
  30.                zebra     5e+03       277     0.176     0.935     0.858     0.297
  31.              giraffe     5e+03       170     0.171     0.959     0.892      0.29
  32.             backpack     5e+03       384    0.0426     0.708     0.392    0.0803
  33.             umbrella     5e+03       392    0.0672     0.878      0.65     0.125
  34.              handbag     5e+03       483    0.0238     0.629     0.242    0.0458
  35.                  tie     5e+03       297    0.0419     0.805     0.599    0.0797
  36.             suitcase     5e+03       310    0.0823     0.855     0.628      0.15
  37.              frisbee     5e+03       109     0.126     0.872     0.796     0.221
  38.                 skis     5e+03       282    0.0473     0.748     0.454     0.089
  39.            snowboard     5e+03        92    0.0579     0.804     0.559     0.108
  40.          sports ball     5e+03       236     0.057     0.733     0.622     0.106
  41.                 kite     5e+03       399     0.087     0.852     0.645     0.158
  42.         baseball bat     5e+03       125    0.0496     0.776     0.603    0.0932
  43.       baseball glove     5e+03       139    0.0511     0.734     0.563    0.0956
  44.           skateboard     5e+03       218    0.0655     0.844      0.73     0.122
  45.            surfboard     5e+03       266    0.0709     0.827     0.651     0.131
  46.        tennis racket     5e+03       183    0.0694     0.858     0.759     0.128
  47.               bottle     5e+03       966    0.0484     0.812     0.513    0.0914
  48.           wine glass     5e+03       366    0.0735     0.738     0.543     0.134
  49.                  cup     5e+03       897    0.0637     0.788     0.538     0.118
  50.                 fork     5e+03       234    0.0411     0.662     0.487    0.0774
  51.                knife     5e+03       291    0.0334     0.557     0.292    0.0631
  52.                spoon     5e+03       253    0.0281     0.621     0.307    0.0537
  53.                 bowl     5e+03       620    0.0624     0.795     0.514     0.116
  54.               banana     5e+03       371     0.052      0.83      0.41    0.0979
  55.                apple     5e+03       158    0.0293     0.741     0.262    0.0564
  56.             sandwich     5e+03       160    0.0913     0.725     0.522     0.162
  57.               orange     5e+03       189    0.0382     0.688      0.32    0.0723
  58.             broccoli     5e+03       332    0.0513      0.88     0.445     0.097
  59.               carrot     5e+03       346    0.0398     0.766     0.362    0.0757
  60.              hot dog     5e+03       164    0.0958     0.646     0.494     0.167
  61.                pizza     5e+03       224    0.0886     0.875     0.699     0.161
  62.                donut     5e+03       237    0.0925     0.827      0.64     0.166
  63.                 cake     5e+03       241    0.0658      0.71     0.539      0.12
  64.                chair     5e+03  1.62e+03    0.0432     0.793     0.489    0.0819
  65.                couch     5e+03       236     0.118     0.801     0.584     0.205
  66.         potted plant     5e+03       431    0.0373     0.852     0.505    0.0714
  67.                  bed     5e+03       195     0.149     0.846     0.693     0.253
  68.         dining table     5e+03       634    0.0546      0.82      0.49     0.102
  69.               toilet     5e+03       179     0.161      0.95      0.81     0.275
  70.                   tv     5e+03       257    0.0922     0.903      0.79     0.167
  71.               laptop     5e+03       237     0.127     0.869     0.744     0.222
  72.                mouse     5e+03        95    0.0648     0.863     0.732      0.12
  73.               remote     5e+03       241    0.0436     0.788     0.535    0.0827
  74.             keyboard     5e+03       117    0.0668     0.923     0.755     0.125
  75.           cell phone     5e+03       291    0.0364     0.704     0.436    0.0692
  76.            microwave     5e+03        88     0.154     0.841     0.743     0.261
  77.                 oven     5e+03       142    0.0618     0.803     0.576     0.115
  78.              toaster     5e+03        11    0.0565     0.636     0.191     0.104
  79.                 sink     5e+03       211    0.0439     0.853     0.544    0.0835
  80.         refrigerator     5e+03       107    0.0791     0.907     0.742     0.145
  81.                 book     5e+03  1.08e+03    0.0399     0.667     0.233    0.0753
  82.                clock     5e+03       292    0.0542     0.836     0.733     0.102
  83.                 vase     5e+03       353    0.0675     0.799     0.591     0.125
  84.             scissors     5e+03        56    0.0397      0.75     0.461    0.0755
  85.           teddy bear     5e+03       245    0.0995     0.882     0.669     0.179
  86.           hair drier     5e+03        11   0.00508    0.0909    0.0475   0.00962
  87.           toothbrush     5e+03        77    0.0371      0.74     0.418    0.0706
  88.  
  89.  Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.409
  90.  Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=100 ] = 0.600
  91.  Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets=100 ] = 0.446
  92.  Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.243
  93.  Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.450
  94.  Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.514
  95.  Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=  1 ] = 0.326
  96.  Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets= 10 ] = 0.536
  97.  Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.593
  98.  Average Recall     (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.422
  99.  Average Recall     (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.640
  100.  Average Recall     (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.707

Reproduce Our Results

This command trains yolov3-spp.cfg from scratch to our mAP above. Training takes about one week on a 2080Ti.

  1. $ python3 train.py --weights '' --cfg yolov3-spp.cfg --epochs 273 --batch 16 --accum 4 --multi --pre

Reproduce Our Environment

To access an up-to-date working environment (with all dependencies including CUDA/CUDNN, Python and PyTorch preinstalled), consider a:

Citation

Contact

Issues should be raised directly in the repository. For additional questions or comments please email Glenn Jocher at glenn.jocher@ultralytics.com or visit us at https://contact.ultralytics.com.

yolov3 in PyTorch的更多相关文章

  1. pytorch实现yolov3(1) yolov3基本原理

    理解一个算法最好的就是实现它,对深度学习也一样,准备跟着https://blog.paperspace.com/how-to-implement-a-yolo-object-detector-in-p ...

  2. pytorch版yolov3训练自己数据集

    目录 1. 环境搭建 2. 数据集构建 3. 训练模型 4. 测试模型 5. 评估模型 6. 可视化 7. 高级进阶-网络结构更改 1. 环境搭建 将github库download下来. git cl ...

  3. [Open Source]基于YOLOv3与Django框架的Web应用-YOLOv3_Detect_Web

    YOLOv3_Detect_Web Use Yolov3 detect on Web 使用 YOLOv3(PyTorch 和 Django 实现)的对象检测应用程序. 网页和 REST API由Dja ...

  4. darknet

    darknet第二作者:https://github.com//AlexeyAB/darknet#how-to-train-to-detect-your-custom-objects yolov3实现 ...

  5. 商汤开源的mmdetection技术报告

    目录 1. 简介 2. 支持的算法 3. 框架与架构 6. 相关链接 前言:让我惊艳的几个库: ultralytics的yolov3,在一众yolov3的pytorch版本实现算法中脱颖而出,收到开发 ...

  6. Python高阶用法总结

    目录 1. lambda匿名函数 1.1 函数式编程 1.2 应用在闭包 2. 列表解析式 3. enumerate内建函数 4. 迭代器与生成器 4.1 迭代器 4.3 生成器 5. 装饰器 前言: ...

  7. 小白也能弄懂的目标检测之YOLO系列 - 第一期

    大家好,上期分享了电脑端几个免费无广告且实用的录屏软件,这期想给大家来讲解YOLO这个算法,从零基础学起,并最终学会YOLOV3的Pytorch实现,并学会自己制作数据集进行模型训练,然后用自己训练好 ...

  8. YOLOv4没交棒，但YOLOv5来了！

    YOLOv4没交棒,但YOLOv5来了! 前言 4月24日,YOLOv4来了! 5月30日,"YOLOv5"来了! 这里的 "YOLOv5" 是带有引号的,因为 ...

  9. Pytorch版本yolov3源码阅读

    目录 Pytorch版本yolov3源码阅读 1. 阅读test.py 1.1 参数解读 1.2 data文件解析 1.3 cfg文件解析 1.4 根据cfg文件创建模块 1.5 YOLOLayer ...

随机推荐

  1. http请求之of_ordering_getmiditem

    //Public function of_ordering_getmiditem (string as_instr,string as_key) returns string //string as_ ...

  2. 北电之死：谁谋杀了华为的对手？——银湖资本（Silver Lake）董事总经理爱德华·詹德出任CEO，既不了解华为，也不重视中国，直截了当地否决了收购华为

    作者:戴老板:微信公众号:饭统戴老板(ID: worldofboss) 2003年5月,北京SARS疫情紧张,摩托罗拉集团总裁迈克·扎菲罗夫斯基(Mike Zafirovski)却准备不走寻常路,决定 ...

  3. vue和react区别

    vue和react区别  

  4. z-index神奇的失效了！！！

    z-index简单介绍 首先z-index只对定位元素有效,什么是定位元素呢,也就是设置了position属性的元素,position:relative--相对定位,position:absolute ...

  5. O059、Backup Volume 操作

    参考https://www.cnblogs.com/CloudMan6/p/5662236.html   BackUp是将Volume备份到别的地方(备份设备),将来可以通过restore操作恢复. ...

  6. vue实现吸顶

    data(){ return{ list:[], swiperOption:"", xiding : "", // 轮播高度 SwiperHeight:'' } ...

  7. php扩展库解释

    扩展库 说明 注解 php_bz2.dll bzip2 压缩函数库 无 php_calendar.dll 历法转换函数库 自 PHP 4.0.3 起内置 php_cpdf.dll ClibPDF 函数 ...

  8. Redis-Hash常用命令

    Redis-Hash常用命令 hset key field value 设置一个散列,但是在散列中一次只能设置一个属性,如果要批量设置多个属性,则需要使用 hmset命令 hget key field ...

  9. idea tomcat 乱码问题的解决及相关设置

    问题,在idea中出现乱码问题,以前没有的,好像在设置系统代码为utf8之后就出现了,于是尝试了一系列办法,希望这些办法对您有帮助. 先看一下乱码的样式. 设置办法 1.在tomcat Server中 ...

  10. 保证在浏览器上word/图片/Excel的下载的表现形式一样

    function downloadImage(src) { console.log(src); //src="http://192.168.12.50:8181/file/common/pn ...