Alpamayo Vision Model v3

Overview

Vision model trained on NVIDIA Physical AI Autonomous Vehicles dataset for trajectory prediction in autonomous driving scenarios.

Training Details

  • Dataset: nvidia/PhysicalAI-Autonomous-Vehicles (streamed)
  • Training steps: 1000
  • Batch size: 4
  • Architecture: CNN (4 conv layers) + GlobalAvgPool + Dense + 3 heads
  • Input: 5-frame sequence, 256x256 RGB, camera_front_wide_120fov
  • Output: trajectory (20x2), lane center offset, heading angle
  • Optimizer: Adam (lr=1e-4)
  • Best trajectory error: 15.74

Usage

Downloads last month
71
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support