Environment building

Environmental Science

• ubuntu 18.04 64bit
• GTX 1070Ti
• anaconda with python 3.8
• pytorch 1.7.1
• cuda 10.1
• yolov5 5.0.9

For ease of use yolov5
object detection , Some netizens have made it into a library , Submitted to the official index library pypi
On , such , We can use it directly pip To install , Its project address : https://github.com/fcakyon/yolov5-pip

install

First create a clean python
A virtual environment

conda create -n yolov5pip python=3.8
conda activate yolov5pip

If python
The version is greater than 3.7, Use it directly pip
install

pip install yolov5

If python
The version is 3.6 Words , You need to install the specified version of numpy
and matplotlib

pip install "numpy>=1.18.5,<1.20" "matplotlib>=3.2.2,<4"
pip install yolov5

Use of the command line

yolov5
After the library is installed , At the same time, it will generate yolov5
Command line tools for , Its use method is the same as that in the source code detect.py
Very similar , as follows

yolov5 detect --source 0 # webcam
file.jpg # image
file.mp4 # video
path/ # directory
path/*.jpg # glob
rtsp://170.93.143.139/rtplive/470011e600ef003a004ee33696235daa # rtsp stream
rtmp://192.168.1.105/live/test # rtmp stream
http://112.50.243.8/PLTV/88888888/224/3221225900/1.m3u8 # http stream

Tests found ,cpu Under the environment of , picture 、 No problem with video detection , But if you use gpu Words , It's going to be a mistake , This may be the current version of bug

YOLOv5 2021-9-15 torch 1.7.1+cu101 CUDA:0 (NVIDIA GeForce GTX 1070 Ti, 8116.4375MB)
Fusing layers…
Model Summary: 283 layers, 7276605 parameters, 0 gradients
Traceback (most recent call last):
File “/home/xugaoxiang/anaconda3/envs/yolov5pip/bin/yolov5”, line 8, in
sys.exit(app())
File “/home/xugaoxiang/anaconda3/envs/yolov5pip/lib/python3.8/site-packages/yolov5/cli.py”, line 11, in app
fire.Fire(
File “/home/xugaoxiang/anaconda3/envs/yolov5pip/lib/python3.8/site-packages/fire/core.py”, line 141, in Fire
component_trace = _Fire(component, args, parsed_flag_args, context, name)
File “/home/xugaoxiang/anaconda3/envs/yolov5pip/lib/python3.8/site-packages/fire/core.py”, line 466, in _Fire
component, remaining_args = _CallAndUpdateTrace(
File “/home/xugaoxiang/anaconda3/envs/yolov5pip/lib/python3.8/site-packages/fire/core.py”, line 681, in _CallAndUpdateTrace
component = fn(*varargs, **kwargs)
File “/home/xugaoxiang/anaconda3/envs/yolov5pip/lib/python3.8/site-packages/torch/autograd/grad_mode.py”, line 26, in decorate_context
return func(*args, **kwargs)
File “/home/xugaoxiang/anaconda3/envs/yolov5pip/lib/python3.8/site-packages/yolov5/detect.py”, line 120, in run
model(torch.zeros(1, 3, *imgsz).to(device).type_as(next(model.parameters()))) # run once
TypeError: zeros() argument after * must be an iterable, not int

except detect Beyond detection ,yolov5 And support train、val and export

The commands for model training are as follows

yolov5 train --data coco.yaml --cfg yolov5s.yaml --weights '' --batch-size 64
                                yolov5m                                40
                                yolov5l                                24
                                yolov5x                                16

On the other val and export Use of commands , Can pass --help To view the specific supported parameters , Such as

yolov5 export --help

stay Python Use in

The library is installed , You can go to python Called in , Let's look at the simplest example

#  The import module 
import yolov5

#  Load model 
model = yolov5.load('yolov5s.pt')

#  Pictures to be tested 
img = 'https://github.com/ultralytics/yolov5/raw/master/data/images/zidane.jpg'

#  Reasoning , Default parameters 
# results = model(img)

#  Use specific dimensions for reasoning 
# results = model(img, size=1280)

#  Data to enhance , Can detect more targets , Of course, there may also be false detection 
results = model(img, augment=True)

#  Analysis of test result data , Category 、 Degree of confidence 、 Target location information 
predictions = results.pred[0]
boxes = predictions[:, :4] # x1, x2, y1, y2
scores = predictions[:, 4]
categories = predictions[:, 5]

#  Show test results 
results.show()

#  Save the test result picture 
results.save(save_dir='results/')

Will be officially provided by yolov5s Download the model , Then execute the above script , The test results are as follows

Due to the use of augment Parameters , There are more test results person, The detection results of the default parameters are the same as the original yolov5 The test results are consistent

Use GPU Accelerated testing

Installation is preferred gpu Version of pytorch, Choose here 1.7.1

pip install torch==1.7.1+cu101 torchvision==0.8.2+cu101 torchaudio==0.7.2 -f https://download.pytorch.org/whl/torch_stable.html

Just like the original , adopt device Parameter to specify cuda Speed up

from yolov5 import YOLOv5

#  Model path 
model_path = 'yolov5s.pt'

device = "cuda" # or "cpu"

#  Model initialization 
yolov5 = YOLOv5(model_path, device)

#  Pictures to be tested 
image = 'https://github.com/ultralytics/yolov5/raw/master/data/images/zidane.jpg'

#  Reasoning , Default parameters 
results = yolov5.predict(image)

#  Use specific dimensions for reasoning 
results = yolov5.predict(image, size=1280)

#  Data to enhance , Can detect more targets , Of course, there may also be false detection 
# results = yolov5.predict(image, augment=True)

#  Use multiple parameters together 
# results = yolov5.predict(image, size=1280, augment=True)

#  If you need to detect multiple pictures , You can use the list 
# results = yolov5.predict([image1, image2], size=1280, augment=True)

#  Analysis of test result data , Category 、 Degree of confidence 、 Target location information 
predictions = results.pred[0]
boxes = predictions[:, :4] # x1, x2, y1, y2
scores = predictions[:, 4]
categories = predictions[:, 5]

#  Show test results 
results.show()

#  Save the test results 
results.save(save_dir='results/')

Use google colab Training YOLOv5 Model

colab What is it?

colab yes google One of the jupyter notebook Tools , Support google drive、tensorflow Inside google Family bucket , It is mainly used in the development and research of machine learning .colab The biggest benefit is to the general public AI Developers offer free gpu resources , It can be run very easily on it, such as tensorflow、pytorch、keras Equal depth learning framework .

YOLOv5 model training

Came to google drive, Click on the top left New

Create a new folder colab

Next , Upload the prepared mask data set to colab In the folder , This dataset , We were in front of YOLOv5 Used in model training , You can download it at the following address

Original link https://public.roboflow.ai/object-detection/mask-wearing

Next create colab, Click on New --> More --> Google Colaboratory

Create good notebook after , Need to come modify --> Notebook settings Set up gpu Speed up

Hardware accelerator , choice GPU, preservation

Click on the Connect , choice Connect to managed code executor .

GPU After setting the environment , We can do that notebook View in colab Provided gpu Resources , Use !nvidia-smi command

You can see google The hardware provided is tesla P100, Explicit memory is 16G. It seems that every time colab The distribution of gpu It's different , Sometimes it's P100, Sometimes it's T4

Let's see below. pytorch Installation of , perform

import torch
torch.__version__

You can see that the platform has been installed by default , And the version is 1.6,CUDA The version is 10.1

If you need to install a third-party library , You can install... Directly in the cell , Such as !pip3 install torchvision

Next, I'll put google drive Mount here , So you can do it colab Use in google drive Resources in

import os
from google.colab import drive
drive.mount('/content/drive')

path = "/content/drive/My Drive"

os.chdir(path)
os.listdir(path)

Execute the code in the above cell , You will be asked to enter the verification code

google drive Will be mounted to the directory /content/drive, You can do it later google drive The files in

The preparations are done , We can download YOLOv5 The source of the , Execute... In cells

git clone https://github.com/ultralytics/yolov5.git

And then switch to google drive, modify yolov5/models/yolov5s.yaml, The original nc: 80 Change it to nc: 2

Then you can come to colab, Enter into yolov5 Catalog , Execute the training command in the cell

python train.py --data ../mask/data.yaml --cfg models/yolov5s.yaml --weights '' --batch-size 64

During training , There is pyyaml An error in the module , This is because pyyaml Why the version is too low , We can solve it by upgrading

pip install -U pyyaml

Keep training