RoyaleTrainer.com
A crowdsourcing platform for Clash Royale image annotations that gathered 82,000+ annotations from 500+ community contributors to support AI training for computer vision models.
The Challenge
Training computer vision models for game analysis requires massive amounts of high-quality annotated data. Traditional data collection methods are expensive, time-consuming, and don't scale well for specialized domains like mobile gaming.
For the Clash Royale AI Fight Vision System, I needed thousands of annotated screenshots to train models for unit detection, card recognition, and tower health estimation - a task that would take months for a single person.
Community-Powered Solution
RoyaleTrainer.com leverages the power of crowdsourcing by creating an engaging web application where community members can contribute annotations while competing on leaderboards and tracking their progress.
Built with Next.js and Supabase, the platform streamlines the annotation process with an intuitive interface, real-time progress tracking, and gamification elements that keep contributors engaged.
The result: a 2,000+ member Discord community that generated over 82,000 annotations, creating one of the largest datasets for Clash Royale computer vision research.
Project Impact
Technical Architecture
The platform uses Next.js for server-side rendering and optimal performance, with Supabase providing real-time database capabilities for tracking annotations and user progress. PostgreSQL stores all annotation data with efficient indexing for fast queries.
The annotation interface features custom-built image tools for precise bounding box creation, with automatic validation and quality control measures. Real-time leaderboards and progress tracking keep contributors motivated and engaged.