How srushtiharyan/asl-to-text works
Project type and purpose
• This is a client-side, single-page web application (“frontend only” static site) whose goal is to let users collect/train their own sign-language gesture dataset in the browser and then run predictions to convert ASL gestures into text.
• The README points out three key features:
• Live capture of sign gestures for custom dataset creation
• On-device training of the model (no server)
• On-device inference/prediction of signs → text
File structure
• Top-level entries:
• index.html – the entry point / main UI
• css/ – styling sheets (mystyle.css, font-awesome)
• js/ – core logic broken into modules:
– main.js (prediction/inference engine)
– train.js (dataset collection & training orchestration)
– tutorial.js (guided walkthrough)
– script.js (UI event wiring)
• docs/, images/, webfonts/ – static assets, docs, tutorials
• No “backend” or database folder—everything runs in the browser.
• External dependencies (FontAwesome, likely TensorFlow.js) are pulled in via CDNs or webfonts folder.
Key insights from README
• It’s deployed as a static site (GitHub Pages / Netlify).
• Emphasis on:
• In-browser dataset creation
• In-browser training
• In-browser inference