clementsan / object_detection_detr_api

AI-driven object detection via FastAPI

Object detection via FastAPI

Aim: AI-driven object detection via FastAPI (on COCO image dataset)

Machine learning models:

• facebook/detr-resnet-50
• facebook/detr-resnet-101

---

Table of contents:

• Direct execution
  • Run FastAPI interface
  • Run API query via curl command
  • Run API query via python script
  • Tests via pytest library
• Execution via docker container
  • Build image and run docker container
  • Run query via API
• Deployment on Docker hub
• MLOps pipeline via GitHub actions

---

1. Direct execution

1.1. Run FastAPI interface

Command line in development mode:

fastapi dev app/main.py

Manual command line in production:

fastapi run app/main.py

Notes:

• Serving at: http://127.0.0.1:8000
• API docs: http://127.0.0.1:8000/docs

1.2. Run API query via curl command

Command lines:

• Endpoint "/":

curl -X 'GET' -H 'accept: application/json' 'http://127.0.0.1:8000/'

• Endpoint "/info":

curl -X 'GET' -H 'accept: application/json' 'http://127.0.0.1:8000/info'

• Endpoint "/api/v1/detect":

curl -X POST -F "image=@./tests/data/savanna.jpg" http://127.0.0.1:8000/api/v1/detect

• Endpoint "/api/v1/detect" with optional model type (e.g. detr-resnet-50 or detr-resnet-101):

curl -X POST -F "image=@./tests/data/savanna.jpg" http://127.0.0.1:8000/api/v1/detect?model=detr-resnet-50

1.3. Run API query via python script

Command line:

python app/inference_api.py -u "http://127.0.0.1:8000/api/v1/detect" -f tests/data/savanna.jpg

1.4. Tests via pytest library

Command lines:

pytest tests/ -v

pytest tests/ -s -o log_cli=true -o log_level=DEBUG

2. Execution via docker container

2.1. Build image and run docker container

Command lines:

sudo docker build -t object-detection-detr-api .

sudo docker run --name object-detection-detr-api-cont -p 8000:8000 object-detection-detr-api

2.2. Run query via API

Command lines:

• Endpoint "/":

curl -X 'GET' 'http://0.0.0.0:8000/' -H 'accept: application/json'

• Endpoint "/info":

curl -X 'GET' -H 'accept: application/json' 'http://0.0.0.0:8000/info'

• Endpoint "/api/v1/detect":

curl -X 'POST' -F "image=@./tests/data/savanna.jpg" http://0.0.0.0:8000/api/v1/detect

• Endpoint "/api/v1/detect" with optional model type (e.g. detr-resnet-50 or detr-resnet-101):

curl -X POST -F "image=@./tests/data/savanna.jpg" http://0.0.0.0:8000/api/v1/detect?model=detr-resnet-50

3. Deployment on Docker hub

This FastAPI application is available as a Docker container on Docker hub

URL: https://hub.docker.com/r/cvachet/object-detection-detr-api

4. MLOps pipeline via GitHub actions

Github actions were created to enable Continuous Integration (CI) and Continuous Deployment (CD) for this FastAPI app.

YAML files:

• Python testing suite: ci_python.yml
• Pushing to docker: publish_docker_image.yaml