MetricAnything: Scaling Metric Depth Pretraining with Noisy Heterogeneous Sources

Dimensions overall score 8.0

• Metric Anything introduces Sparse Metric Prompt, a universal interface that decouples spatial reasoning from sensor and camera biases by randomly masking depth maps.method 90%
  Evidencepartial

  Central to our approach is the Sparse Metric Prompt, created by randomly masking depth maps, which serves as a universal interface that decouples spatial reasoning from sensor and camera biases.

  Implicationpartial

  Directly stated in abstract as central to the approach with clear description of mechanism and purpose

  Verificationpartial

  partial

• Metric Anything demonstrates for the first time a clear scaling trend in the metric depth track using about 20M image-depth pairs across 10000 camera models.result 90%
  Evidencepartial

  Using about 20M image-depth pairs spanning reconstructed, captured, and rendered 3D data across 10000 camera models, we demonstrate-for the first time-a clear scaling trend in the metric depth track.

  Implicationpartial

  Explicitly stated with specific numerical data and 'for the first time' qualification

  Verificationpartial

  partial

• The pretrained model excels at prompt-driven tasks including depth completion, super-resolution and Radar-camera fusion.result 85%
  Evidencepartial

  The pretrained model excels at prompt-driven tasks such as depth completion, super-resolution and Radar-camera fusion

  Implicationpartial

  Directly stated in abstract with specific task examples

  Verificationpartial

  partial

• The distilled prompt-free student achieves state-of-the-art results on monocular depth estimation, camera intrinsics recovery, single/multi-view metric 3D reconstruction, and VLA planning.result 85%
  Evidencepartial

  its distilled prompt-free student achieves state-of-the-art results on monocular depth estimation, camera intrinsics recovery, single/multi-view metric 3D reconstruction, and VLA planning.

  Implicationpartial

  Directly stated with specific task list and 'state-of-the-art' claim

  Verificationpartial

  partial

• Using pretrained ViT of Metric Anything as a visual encoder significantly boosts Multimodal Large Language Model capabilities in spatial intelligence.result 80%
  Evidencepartial

  We also show that using pretrained ViT of Metric Anything as a visual encoder significantly boosts Multimodal Large Language Model capabilities in spatial intelligence.

  Implicationpartial

  Directly stated but without specific quantitative evidence in the provided text

  Verificationpartial

  partial

• Metric depth estimation can benefit from the same scaling laws that drive modern foundation models.technical 85%
  Evidencepartial

  These results show that metric depth estimation can benefit from the same scaling laws that drive modern foundation models

  Implicationpartial

  Direct conclusion stated in abstract based on demonstrated results

  Verificationpartial

  partial

• The framework learns metric depth from noisy, diverse 3D sources without manually engineered prompts, camera-specific modeling, or task-specific architectures.method 90%
  Evidencepartial

  a simple and scalable pretraining framework that learns metric depth from noisy, diverse 3D sources without manually engineered prompts, camera-specific modeling, or task-specific architectures.

  Implicationpartial

  Explicitly stated in abstract with clear description of what the approach avoids

  Verificationpartial

  partial

• Scaling metric depth estimation remains challenging due to heterogeneous sensor noise, camera-dependent biases, and metric ambiguity in noisy cross-source 3D data.technical 90%
  Evidencepartial

  Scaling has powered recent advances in vision foundation models, yet extending this paradigm to metric depth estimation remains challenging due to heterogeneous sensor noise, camera-dependent biases, and metric ambiguity in noisy cross-source 3D data.

  Implicationpartial

  Direct statement of the problem being addressed in the abstract

  Verificationpartial

  partial

Competitive landscape

MetricAnything provides a scalable pretraining framework for metric depth estimation from diverse 3D data sources, achieving state-of-the-art results and broadening AI applications in spatial intelligence.