Feature Space Embedding

Embedding Model:

Comparing detections from real-trained (blue) vs. Synetic-trained (orange) models. Filter shows detections close to human ground truth.

💡 Click legend items to show/hide models

🖱️ Click any point to view detection details

🎚️ Slider filters by distance to human annotations

Real-Trained Model

Synetic-Trained Model

Comparison Mode:

Show only detections within threshold distance of human-annotated ground truth (comparing across all images)

Distance Threshold: 6.0

Proof: Synthetic training data creates superior feature representations

100% Synetic-trained detections within threshold of ground truth

100% Real-trained detections within threshold of ground truth