Skip to main content

Welcome To My

Portfolio Website

www.shafayatsaad.vercel
Back
ProjectsTACTIC-FP Annotator

TACTIC-FP Annotator

TACTIC-FP Annotator is a production-grade, keyboard-first web platform for labeling football (soccer) tactical intent at the clip level. It produces clean, benchmark-ready datasets for the TACTIC-Bench research framework — with live video streaming, team-aware possession logic, automatic segment creation, and one-click export. Purpose-built for the TACTIC-FP / TACTIC-Bench research framework, it delivers a fast, distraction-free, single-screen annotation experience. Manual annotation of football tactical intent is highly error-prone, slow, and lacks standardized interfaces or quality gates to ensure dataset reliability. Built a full-stack Next.js web application integrating native HTML5 range-request streaming, custom keyboard shortcut routing, and a Python-FFmpeg background processing pipeline. Enables researchers and annotators to generate benchmark-ready datasets for the TACTIC-Bench framework with high precision, speed, and automated validation rules.

7

Tech Stack

6

Key Features

GitHub

Technologies Used

Next.js 14
React 18
TypeScript
Tailwind CSS
Python 3
FFmpeg
OpenCV
TACTIC-FP Annotator - image 1

Key Features

  • Keyboard-Driven: Every action — intents, navigation, segment creation, submit — has a single-key shortcut for high throughput.
  • Auto-Segmentation: Press 'O' at any playhead position to instantly create a timed segment, with auto-splitting for segments > 15 seconds.
  • Possession-Aware Logic: Attack intents are automatically disabled/enabled based on which team has the ball, with manual override available.
  • FFmpeg Asynchronous Pipeline: Runs background MKV to MP4 video conversion with a live progress bar, reading precise video metadata via ffprobe.
  • Quality Gate: Blocks annotation submissions unless validation rules are met (e.g., tracked players >= 18, coverage estimate >= 80%).
  • Research-Ready Exports: One-click export of structured annotations to JSON and CSV formats matching the TACTIC-Bench schema.