DETRfastapi

Object Detection API powered by DETR and FastAPI

How to run the prediciton API:

A. Using Docker [recommended]:

1. Build Docker image from the Dockerfile included in the repo.

   cd fastapi
   docker build -t fastapi .
   

2. Run the Docker image in detached mode.

   docker run --rm -it --name fastapicontainer -p 90:80 fastapi
   

Now the API endpoint should be live at http://127.0.0.1:90

B. Using venv / normal python environment:

1. Install the necessary python packages.

   cd fastapi
   pip install -r requirements.txt
   

2. Start the server.

   cd fastapi/app
   uvicorn main:app --reload
   

Now the API endpoint should be live at http://127.0.0.1:8000

How to store big model files using DVC + Azure:

1. Initialize dvc repository:

   dvc init
   

2. Add .pth files under DVC version control:

   dvc add fastapi/resnet50-19c8e357.pth
   git add fastapi/resnet50-19c8e357.pth.dvc fastapi/.gitignore
   dvc add fastapi/app/detr-r50-e632da11.pth
   git add fastapi/app/detr-r50-e632da11.pth.dvc fastapi/app/.gitignore
   

3. Set up remote Azure Cloud:

   dvc remote add -d fastapi azure://fastapi/storage
   dvc remote modify --local fastapi connection_string "XXXXX" # Get XXXXX from the proper Azure BlobStorage account. [in this case, nvi0426asa -> Access Keys] 
   

4. Push the data files in Azure Cloud:

   dvc push fastapi/resnet50-19c8e357.pth.dvc
   dvc push fastapi/app/detr-r50-e632da11.pth.dvc
   

Now these 2 model files will be tracked by DVC and can be accessed any time by performing dvc pull fastapi/resnet50-19c8e357.pth.dvc and dvc pull fastapi/app/detr-r50-e632da11.pth.dvc