Hasty Briefsbeta

Bilingual

DepthART: Scaling Foundation Monocular Depth to Tiny Models

13 hours ago
  • Recent geometric foundation models have improved monocular depth estimation, but their benefits are limited for tiny models.
  • DepthART is a compact model designed for robust on-device depth estimation across diverse scenes.
  • It uses bias-resistant data sampling and camera-conditioned fine-tuning to address dataset-specific overfitting and unstable metric adaptation.
  • These methods improve cross-dataset generalization and metric depth prediction in capacity-constrained models.
  • DepthART has been accepted to ACM Multimedia 2026.