Unverified50% confidenceFactExact time
Depth Anything、MiDaS、DPT等基础模型大幅提升了相对深度质量，但输出仍为无尺度的相对值，需要额外锚定手段还原绝对度量
1
Sources
50%
Confidence
Long-term
Relevance
7/11/2026
First Seen
Sources
Road Damage Measurement Using Only a GoPro: A Low-Cost Inspection Solution with Monocular Cameras + AI
redditr/computervision7/10/2026
Related Claims
UnverifiedMonocular depth estimation outputs relative depth rather than absolute metric depth, which is an inherent limitation of this technical approach.74% similarUnverified单目深度估计的输出为相对深度而非绝对度量深度，Depth Anything V2采用仿射不变深度训练范式72% similarUnverified单目深度估计的代表性模型包括Intel的MiDaS、Meta的DPT以及Depth Anything系列71% similarUnverified深度ControlNet接受由MiDaS、DPT或Depth Anything等模型估算的深度图作为输入71% similarUnverified深度重绘使用MiDaS、DPT等单目深度估计模型从原图中自动推断深度图68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/486879API
curl https://kongchang.com/api/v1/knowledge/claims/486879MCP
get_claim(id=486879)