Home

FlashTrack Documentation

Welcome to the FlashTrack documentation — an ultra-lightweight real-time multi-object tracking framework.

Overview

FlashTrack is part of the FlashVision family. It combines a lightweight ReID (re-identification) feature extractor built on ShuffleNetV2 with Kalman filtering for robust multi-object tracking.

Documentation

• Installation
• Quick Start
• Models
• Training
• Trackers
• LoRA Fine-Tuning
• FAQ

Architecture

Input Crop → ShuffleNetV2 → Feature Encoder → ReID Head → Embedding
                                                        └→ Logits (training)

Model Sizes

Model	Params	FP16 Size	Use Case
m-0.5x	~0.3M	~0.6 MB	Ultra-edge (MCU, mobile)
m	~0.8M	~1.6 MB	General edge
m-1.5x	~1.5M	~3.0 MB	High accuracy