FlowerPower

Classifying flowers using the Flower Dataset

A tiny project for Complori

Install

• Clone repository

  git clone https://github.com/sycz00/FlowerPower.git
  

• Create Docker image

  docker build -t flower .
  

Docker Networking and Running

• Assign a static IP to container by defining subnetmasking

docker network create --subnet 192.0.2.0/24 flower_power_net

• Run Docker image with static ip (remove -d flag if you don't want to run the container in the background)

docker run -p 5000:80 -itd --network=flower_power_net --ip=192.0.2.69 flower

• Find and Stop Docker service when using -d flag

docker ps

docker stop <CONTAINER ID>

Following the instructions until this point, the Docker container should be running

Using the API

• Inference of example image daisy.png by sending it to static ip assigned in previous step

curl -X POST -F "file=@inference_test/daisy.jpg" http://192.0.2.69:5000/predict

• Request a new training process, includes overwriting the existing checkpoint with new one. Custom number of epochs and learning-rate can be used here

curl -X POST -H "Content-Type: application/json" -d '{"epochs": 10,"lr":0.001}' http://192.0.2.69:5000/train

Process

• The following image shows the training progress on the validation dataset. it can be seen that the shallow CNN makes reasonable progress.

Dataset

• Flower Dataset