0.3B Student Model Runs Edge Depth Estimation

thetripathi58 · x · 2026-07-08

The post states that after being distilled into a 0.3B student model, the architecture achieves a depth estimation RMSE of 0.310. At one-third the size of NVIDIA's 1B AM-RADIO, it still outperforms the latter.

The author points out that this enables the model to run directly on edge devices with lower latency.

Related event: Ant Robbyant Open-Sources LingBot Vision Models, Topping Depth Benchmarks(18 posts)→

Original post →

More from Embodied

Embodied channel →