AC215_AutoCap/
├── deployment/
│ ├── scripts/ # includes scripts for docker image deployment and K8s cluster creation
│ ├── Dockerfile
│ ├── docker-shell.sh
│ └── README.md # setup deployment config and steps to deploy app to GCP
├── extension/ # browser extension files
│ └── safari/
├── models
├── notebooks
├── references
├── src/ # main source code
│ ├── api/
│ ├── frontend/
│ └── README.md # instructions for local development
├── submissions/
│ ├── milestone1_AutoCap/
│ ├── milestone2_AutoCap/
│ └── milestone3_AutoCap/
├── .gitignore
├── LICENSE
└── README.md
These components are located within src.
- API - Flask API served with
gunicorn. Our API exposes two endpoints,/modelsand/predict. A GET request to/modelsreturns a JSON object which includes all of the vision and language models that are available. A POST request to/predictthat includes image data and specific vision and language models will return an image caption and an attention array. - Frontend - Single-page React application. Static files are served with
nginx. Main components of this app include an image uploader, attention overlay, and caption. More specifically, our image uploader will make a POST request to our/predictAPI endpoint with the vision and language models selected in the settings pane. The response of this request will update the caption component. Further, hovering over each word within the caption will update the attention overlay component that sits above the image.
Located within extension/chrome.
The AutoCap Chrome extension interfaces with the backend server API to iterate through all images without an alt attribute on a webpage, upload these images, and generate image captions. Return captions are added to the corresponding img tag. Hopefully, this will allow people with visual impairments that heavily rely on screen readers to better understand the content of images on websites they visit. We also plan to create a similar Safari extension in the future.