Introduction

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.

Description

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/.

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

Jupyter Notebook

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

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.

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:

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

$ git clone https://github.com/ultralytics/yolov3 && cd yolov3

# convert darknet cfg/weights to pytorch model
$ python3  -c "from models import *; convert('cfg/yolov3-spp.cfg', 'weights/yolov3-spp.weights')" Success: converted 'weights/yolov3-spp.weights' to 'converted.pt' # convert cfg/pytorch model to darknet weights $ python3 -c "from models import *; convert('cfg/yolov3-spp.cfg', 'weights/yolov3-spp.pt')" Success: converted 'weights/yolov3-spp.pt' to 'converted.weights'

mAP

python3 test.py --weights ... --cfg ...

[email protected] run at --nms-thres 0.5, [email protected] 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

$ python3 test.py --save-json --img-size 608 --nms-thres 0.5 --weights ultralytics68.pt

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') Using CUDA device0 _CudaDeviceProperties(name='GeForce RTX 2080 Ti', total_memory=11019MB) Class Images Targets P R [email protected] F1: 100%|███████████████████████████████████████████████████████████████████████████████████| 313/313 [09:46<00:00, 1.09it/s] all 5e+03 3.58e+04 0.0823 0.798 0.595 0.145 person 5e+03 1.09e+04 0.0999 0.903 0.771 0.18 bicycle 5e+03 316 0.0491 0.782 0.56 0.0925 car 5e+03 1.67e+03 0.0552 0.845 0.646 0.104 motorcycle 5e+03 391 0.11 0.847 0.704 0.194 airplane 5e+03 131 0.099 0.947 0.878 0.179 bus 5e+03 261 0.142 0.874 0.825 0.244 train 5e+03 212 0.152 0.863 0.806 0.258 truck 5e+03 352 0.0849 0.682 0.514 0.151 boat 5e+03 475 0.0498 0.787 0.504 0.0937 traffic light 5e+03 516 0.0304 0.752 0.516 0.0584 fire hydrant 5e+03 83 0.144 0.916 0.882 0.248 stop sign 5e+03 84 0.0833 0.917 0.809 0.153 parking meter 5e+03 59 0.0607 0.695 0.611 0.112 bench 5e+03 473 0.0294 0.685 0.363 0.0564 bird 5e+03 469 0.0521 0.716 0.524 0.0972 cat 5e+03 195 0.252 0.908 0.78 0.395 dog 5e+03 223 0.192 0.883 0.829 0.315 horse 5e+03 305 0.121 0.911 0.843 0.214 sheep 5e+03 321 0.114 0.854 0.724 0.201 cow 5e+03 384 0.105 0.849 0.695 0.187 elephant 5e+03 284 0.184 0.944 0.912 0.308 bear 5e+03 53 0.358 0.925 0.875 0.516 zebra 5e+03 277 0.176 0.935 0.858 0.297 giraffe 5e+03 170 0.171 0.959 0.892 0.29 backpack 5e+03 384 0.0426 0.708 0.392 0.0803 umbrella 5e+03 392 0.0672 0.878 0.65 0.125 handbag 5e+03 483 0.0238 0.629 0.242 0.0458 tie 5e+03 297 0.0419 0.805 0.599 0.0797 suitcase 5e+03 310 0.0823 0.855 0.628 0.15 frisbee 5e+03 109 0.126 0.872 0.796 0.221 skis 5e+03 282 0.0473 0.748 0.454 0.089 snowboard 5e+03 92 0.0579 0.804 0.559 0.108 sports ball 5e+03 236 0.057 0.733 0.622 0.106 kite 5e+03 399 0.087 0.852 0.645 0.158 baseball bat 5e+03 125 0.0496 0.776 0.603 0.0932 baseball glove 5e+03 139 0.0511 0.734 0.563 0.0956 skateboard 5e+03 218 0.0655 0.844 0.73 0.122 surfboard 5e+03 266 0.0709 0.827 0.651 0.131 tennis racket 5e+03 183 0.0694 0.858 0.759 0.128 bottle 5e+03 966 0.0484 0.812 0.513 0.0914 wine glass 5e+03 366 0.0735 0.738 0.543 0.134 cup 5e+03 897 0.0637 0.788 0.538 0.118 fork 5e+03 234 0.0411 0.662 0.487 0.0774 knife 5e+03 291 0.0334 0.557 0.292 0.0631 spoon 5e+03 253 0.0281 0.621 0.307 0.0537 bowl 5e+03 620 0.0624 0.795 0.514 0.116 banana 5e+03 371 0.052 0.83 0.41 0.0979 apple 5e+03 158 0.0293 0.741 0.262 0.0564 sandwich 5e+03 160 0.0913 0.725 0.522 0.162 orange 5e+03 189 0.0382 0.688 0.32 0.0723 broccoli 5e+03 332 0.0513 0.88 0.445 0.097 carrot 5e+03 346 0.0398 0.766 0.362 0.0757 hot dog 5e+03 164 0.0958 0.646 0.494 0.167 pizza 5e+03 224 0.0886 0.875 0.699 0.161 donut 5e+03 237 0.0925 0.827 0.64 0.166 cake 5e+03 241 0.0658 0.71 0.539 0.12 chair 5e+03 1.62e+03 0.0432 0.793 0.489 0.0819 couch 5e+03 236 0.118 0.801 0.584 0.205 potted plant 5e+03 431 0.0373 0.852 0.505 0.0714 bed 5e+03 195 0.149 0.846 0.693 0.253 dining table 5e+03 634 0.0546 0.82 0.49 0.102 toilet 5e+03 179 0.161 0.95 0.81 0.275 tv 5e+03 257 0.0922 0.903 0.79 0.167 laptop 5e+03 237 0.127 0.869 0.744 0.222 mouse 5e+03 95 0.0648 0.863 0.732 0.12 remote 5e+03 241 0.0436 0.788 0.535 0.0827 keyboard 5e+03 117 0.0668 0.923 0.755 0.125 cell phone 5e+03 291 0.0364 0.704 0.436 0.0692 microwave 5e+03 88 0.154 0.841 0.743 0.261 oven 5e+03 142 0.0618 0.803 0.576 0.115 toaster 5e+03 11 0.0565 0.636 0.191 0.104 sink 5e+03 211 0.0439 0.853 0.544 0.0835 refrigerator 5e+03 107 0.0791 0.907 0.742 0.145 book 5e+03 1.08e+03 0.0399 0.667 0.233 0.0753 clock 5e+03 292 0.0542 0.836 0.733 0.102 vase 5e+03 353 0.0675 0.799 0.591 0.125 scissors 5e+03 56 0.0397 0.75 0.461 0.0755 teddy bear 5e+03 245 0.0995 0.882 0.669 0.179 hair drier 5e+03 11 0.00508 0.0909 0.0475 0.00962 toothbrush 5e+03 77 0.0371 0.74 0.418 0.0706 Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.409 Average Precision (AP) @[ IoU=0.50 | area= all | maxDets=100 ] = 0.600 Average Precision (AP) @[ IoU=0.75 | area= all | maxDets=100 ] = 0.446 Average Precision (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.243 Average Precision (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.450 Average Precision (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.514 Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 1 ] = 0.326 Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 10 ] = 0.536 Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.593 Average Recall (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.422 Average Recall (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.640 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.

$ 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:

GCP Deep Learning VM with $300 free credit offer: See our GCP Quickstart Guide
Google Colab Notebook with 12 hours of free GPU time: Google Colab Notebook
Docker Image from https://hub.docker.com/r/ultralytics/yolov3. See Docker Quickstart Guide

Citation

Contact

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

yolov3 in PyTorch