Object Detection App based on Flask in Docker

This a simple Object Detection App working on browser using Flask and YOLOv3. The user can select any image and the app will visualise an image with results together with a list of detected objects and a counter showing the total number of deteted objects.

The object detection based on YOLOv3

• Model Used

YOLOv3 320x320 from Darknet Weights are available at YOLOv3 Weights

The configuration file used from cfg File The labels for COCO are used from coco names

Note Download the model weights and save it under data/model/ as yolov3.weights before running the App

Dependencies

Making the app run locally, the dependencies shall be created in a virtual enviorment as follows:

Using on Anaconda

Create an enviorment for the app using

conda env create -f environment.yml

The complete enviorment is listed under the file environment_all.yml

Using on virtualenvs

Create the virtual enviorment and install the required libraries from requirements.txt

pip install -r requirements.txt

API Calls

• /show_image

Used to visualise the uploaded image

• /detect_object

Used to perform object detection using YOLOv3

Running the App

Run the app in Debug Mode

flask --app flask_object_detection_app --debug run

OR

python flask_object_detection_app.py

Running the App in Docker

1. Build the Docker Image

Build the image with docker build -t odapp .

2. Run the app in Docker After sucessful building, run the docker docker run -ti --rm -p 5000:5000 odapp

and open a browser in the host machine http://localhost:5000/ to visualise the app.

App Views:

Page1: Landing page

You can select the image in this page

Page 2 Confirmation message on sucessful upload

Page 3 Visualise the uploaded image

Page 4: Perform Object Detection, counting and visualise the results

Reference

Developed from the tutorial from Thinkinfi