DepthART: A 6M-Parameter Depth Model Keeps Foundation-Model Generalization on Edge Devices
jiqizhixin · x · 2026-10-10
DepthART (accepted at ACM Multimedia 2026) shows a tiny 6M-parameter model can retain the cross-scene generalization that makes depth foundation models like Depth Anything, Metric3D and UniDepth useful — small enough to run on a Jetson edge device. The trick: tiny models' limited capacity gets dominated by high-frequency data, so the team started from 44M multi-source candidate images and down-sampled dense, repetitive samples in visual feature space while preserving sparse regions, yielding 1.7M more balanced training images.
More from Embodied
- "She's a 10/10, but she's a robot": humanoid demo turns heads — creatoroff · 2026-10-10
- Xpeng shows Physical AI at Paris Motor Show, vision-only FSD heading to Europe — LinusEkenstam · 2026-10-10
- FIND preprint: robot picks its own weaknesses, lifts 8-task success from 55% to 71.9% — GeorgiaChal · 2026-10-10
- Musk: Digital Optimus beats Diablo halfway through with no APIs, just screen pixels — XFreeze · 2026-10-10
- Tongji team builds spiderweb-inspired 6-axis robotic 3D printer with self-supporting structures — lukas_m_ziegler · 2026-10-10
- Robot hand turns book pages: UVTA visual-tactile-action model hits 70% vs 29% baseline — CyberRobooo · 2026-10-10