Recyclops Backend

The backend which serves the pytorch model for image inference

Technologies

Tech	Version
Python	3.9.5
Flask	2.1.2
PyTorch	1.12.0

Usage

The server only exposes a single POST request on {hostname}/predict

Using Curl

curl -X POST -F file=@"<path to img>" {hostname}/predict

Setup (Local)

This setup was only tested on python 3.9.5

1. Install python dependencies

pip install -r requirements.txt

2. (Optional) edit model & label maps in app.py

MODEL_PATH = 'models/resnet_large_resize_150_cpu.model'
LABEL_MAP_PATH = 'label_maps/full_label_map.json'

Note: Label maps must correspond to the model used (specified in the file name)

3. Run server

flask run

Deployment (Heroku)

Do use the heroku branch if you're following this steps

1. Setup new heroku

heroku login -i
heroku create <app name>

2. Setup heroku remote and push

heroku git:remote -a <app name>
git push heroku heroku:master

3. Setup environments for model & label map

heroku config:set MODEL_PATH=models/resnet_large_resize_150_cpu.model
heroku config:set LABEL_MAP_PATH=label_maps/full_label_map.json

The server should be up by now

Note: Feel free to change the model path and label map path, however as mentioned above do make sure they correspond to each other