ApexTrack AI — Track Condition Classifier V2
ApexTrack AI is a computer vision model for live racing track condition classification (dry, damp, wet). It powers dynamic strategy decision-support systems for race engineers and simulation platforms.
Model Overview
- Architecture: Vision Transformer (ViT-Base,
google/vit-base-patch16-224-in21k) - Task: 3-Class Image Classification (
dry,damp,wet) - Input Resolution: 224x224 RGB images
- Labels:
0:damp1:dry2:wet
Dataset & Training
- Source Dataset: Weather Whiplash Surfaces (Real road & asphalt track surface conditions)
- Dataset Size: 190 original annotated images split into train (131), validation (27), and test (32).
- V2 Balancing: Controlled, conservative image augmentation was applied ONLY to the training split (bringing each training class to exactly 100 images = 300 total training images).
- Validation & Test Sets: Remained 100% untouched and unaugmented to guarantee zero data leakage and honest evaluation.
Performance Metrics (Evaluated on Untouched Test Set)
- Accuracy: 43.75%
- Macro F1: 40.78%
- Weighted F1: 42.51%
Per-Class Performance
| Class | Precision | Recall | F1-Score |
|---|---|---|---|
| Dry | 50.00% | 72.73% | 59.26% |
| Damp | 22.22% | 28.57% | 25.00% |
| Wet | 57.14% | 28.57% | 38.10% |
Note on "Drying" Condition
"Drying" is not an image classification class. Instead, track drying is inferred temporally by the ApexTrack AI backend engine across sequential live predictions (e.g. wet → damp → dry).
Limitations & Disclaimer
- Prototype Status: This is an educational/hackathon prototype trained on a compact dataset.
- Decision Support: Predictions are designed for advisory decision support and should not be used in safety-critical autonomous control systems without human verification.