This repo shows you how to deploy a model trained in FastAI to a web API using only PyTorch. First follow the given notebook to remove the FastAI dependency.
The repo comes with a model trained to recognise 5 types of traffic signs as an example and this model was trained using FastAI.
To run the website locally use a command line tool to navigate to the folder where you downloaded or cloned this repo, then type:
python app.py
This will start up a Flask server that you can use via a UI or via an API.
To verify everything is working run:
python tests.py
To change the model to yours, change the static\pytorch_traffic_sign file with your own pytorch model.
If you are using a model originally trained with FastAI, make sure to verify that you have replaced the head of the model properly using the instructions in the Google Colab notebook.
<br>
After changing the model, also create a json file with the names of all categories your trained on. For reference check static\traffic_class.json
The app and API requires PyTorch, TorchVision and Flask only, it has been tested with the following version
- Flask==1.0.3
- PyTorch==1.3.0
- Torchvision==0.4.0
The code here is based on the PyTorch documentation here as well as a linked repository here.