AI Deploy - Tutorial - Deploy a web service for YOLOv5 using Flask

Objective

The purpose of this tutorial is to show how to deploy a web service for YOLOv5 using your own weights generated after training a YOLOv5 model on your dataset.

In order to do this, you will use Flask, an open-source micro framework for web development in Python. You will also learn how to build and use a custom Docker image for a Flask application.

For more information on how to train YOLOv5 on a custom dataset, refer to the following documentation.

Requirements

• access to the OVHcloud Control Panel
• an AI Deploy project created inside a Public Cloud project
• a user for AI Deploy
• Docker installed on your local computer
• some knowledge about building image and Dockerfile
• your weights obtained from training a YOLOv5 model on your dataset (refer to the "Export trained weights for future inference" part of the notebook for YOLOv5)

Instructions

First, the tree structure of your folder should be as follows.

• You have to create the folder named models_train and this is where you can store the weights generated after your trainings. You are free to put as many weight files as you want in the models_train folder.

• Here we will mainly discuss how to write the app.py code, the requirements.txt file and the Dockerfile. If you want to see the whole code, please refer to the GitHub repository.

Write the Flask application

Create a Python file named app.py.

Inside that file, import your required modules:

import sys
import io
from PIL import Image
import cv2
import torch
from flask import Flask, render_template, request, make_response
from werkzeug.exceptions import BadRequest
import os

Create Flask app:

app = Flask(__name__)

Load your own weights:

Here a python dictionary is created to store the name of each of your weight files (key) and the corresponding model (value).

# create a python dictionary for your models d = {<key>: <value>, <key>: <value>, ..., <key>: <value>}
dictOfModels = {}
# create a list of keys to use them in the select part of the html code
listOfKeys = []

Write the inference function:

def get_prediction(img_bytes,model):
    img = Image.open(io.BytesIO(img_bytes))
    # inference
    results = model(img, size=640)  
    return results

Define the GET method:

@app.route('/', methods=['GET'])
def get():
  # in the select we will have each key of the list in option
  return render_template("index.html", len = len(listOfKeys), listOfKeys = listOfKeys)

Define the POST method:

@app.route('/', methods=['POST'])
def predict():
    file = extract_img(request)
    img_bytes = file.read()
    # choice of the model
    results = get_prediction(img_bytes,dictOfModels[request.form.get("model_choice")])
    print(f'User selected model : {request.form.get("model_choice")}')
    # updates results.imgs with boxes and labels
    results.render()
    # encoding the resulting image and return it
    for img in results.ims:
        RGB_img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
        im_arr = cv2.imencode('.jpg', RGB_img)[1]
        response = make_response(im_arr.tobytes())
        response.headers['Content-Type'] = 'image/jpeg'
    return response

def extract_img(request):
    # checking if image uploaded is valid
    if 'file' not in request.files:
        raise BadRequest("Missing file parameter!")
    file = request.files['file']
    if file.filename == '':
        raise BadRequest("Given file is invalid")
    return file

Define the main and start your app:

if __name__ == '__main__':

    print('Starting yolov5 webservice...')
    # Getting directory containing models from command args (or default 'models_train')
    models_directory = 'models_train'
    if len(sys.argv) > 1:
        models_directory = sys.argv[1]
    print(f'Watching for yolov5 models under {models_directory}...')
    for r, d, f in os.walk(models_directory):
        for file in f:
            if ".pt" in file:
                # example: file = "model1.pt"
                # the path of each model: os.path.join(r, file)
                model_name = os.path.splitext(file)[0]
                model_path = os.path.join(r, file)
                print(f'Loading model {model_path} with path {model_path}...')
                dictOfModels[model_name] = torch.hub.load('ultralytics/yolov5', 'custom', path=model_path, force_reload=True)
                # you would obtain: dictOfModels = {"model1" : model1 , etc}
        for key in dictOfModels :
            listOfKeys.append(key) # put all the keys in the listOfKeys

    # starting app
    app.run(debug=True,host='0.0.0.0')

Find more information about the Flask application here to get ready to use it.

Write the requirements.txt file for the application

The requirements.txt file will allow us to write all the modules needed to make our application work. This file will be useful when writing the Dockerfile.

Flask==2.1.0
torch==1.13.1
torchvision==0.14.1
psutil==5.9.4
IPython==8.8.0
requests==2.28.2
PyYAML==6.0
tqdm==4.64.1
pandas==1.5.2
opencv-python-headless==4.6.0.66
matplotlib==3.6.3
seaborn==0.12.2

Write the Dockerfile for the application

Your Dockerfile should start with the FROM instruction indicating the parent image to use. In our case we choose to start from a python:3.8 image:

FROM python:3.8

Create the home directory and add your files to it:

WORKDIR /workspace
ADD . /workspace

Install the requirements.txt file which contains your needed Python modules using a pip install ... command:

RUN pip install -r requirements.txt

Define your default launching command to start the application:

CMD [ "python" , "/workspace/app.py" ]

Give correct access rights to ovhcloud user (42420:42420):

RUN chown -R 42420:42420 /workspace
ENV HOME=/workspace

Build the Docker image from the Dockerfile

From the directory containing your Dockerfile, run one of the following commands to build your application image:

# Build the image using your machine's default architecture
docker build . -t flask-yolov5:latest

# Build image targeting the linux/amd64 architecture
docker buildx build --platform linux/amd64 -t flask-yolov5:latest .

• The first command builds the image using your system’s default architecture. This may work if your machine already uses the linux/amd64 architecture, which is required to run containers with our AI products. However, on systems with a different architecture (e.g. ARM64 on Apple Silicon), the resulting image will not be compatible and cannot be deployed.

• The second command explicitly targets the linux/AMD64 architecture to ensure compatibility with our AI services. This requires buildx, which is not installed by default. If you haven’t used buildx before, you can install it by running: docker buildx install

Test it locally (optional)

Launch the following Docker command to launch your application locally on your computer:

docker run --rm -it -p 5000:5000 --user=42420:42420 flask-yolov5:latest

Once started, your application should be available on http://localhost:5000.

Push the image into the shared registry

Find the address of your shared registry by launching this command:

ovhai registry list

Login on the shared registry with your usual AI Platform user credentials:

docker login -u <user> -p <password> <shared-registry-address>

Push the compiled image into the shared registry:

docker tag flask-yolov5:latest <shared-registry-address>/flask-yolov5:latest
docker push <shared-registry-address>/flask-yolov5:latest

Launch the AI Deploy app

The following command starts a new app running your Flask application:

ovhai app run --default-http-port 5000 --cpu 4 <shared-registry-address>/flask-yolov5:latest

Go further

• You can imagine deploying a Flask app in order to classify the feelings in a text. Refer to this tutorial.
• Another way to create an AI Deploy app is to use Streamlit! Here it is.

If you need training or technical assistance to implement our solutions, contact your sales representative or click on this link to get a quote and ask our Professional Services experts for a custom analysis of your project.

Feedback

Please send us your questions, feedback and suggestions to improve the service:

• On the OVHcloud Discord server