localai-org/depth-anything.cpp

Health v2 · maintenance only

Flags: no_releases young

Adoption not part of the score

1090 stars · 110 forks observed · 2026-09-01

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

A from-scratch C++17/ggml port of ByteDance's Depth Anything 2 and 3 models for dependency-free monocular metric depth and camera pose inference. It ships as a self-contained GGUF-based native library and CLI with a flat C API, running faster than PyTorch on CPU with optional CUDA/Metal/Vulkan backends.

Use cases

• estimate metric depth maps from a single image without Python or PyTorch
• recover camera pose and intrinsics from photos for 3D reconstruction
• export depth and point clouds to glb, COLMAP, or PLY formats
• embed depth estimation into C, C++, Go, or Rust applications via a C API
• run quantized depth models on CPU with no GPU or CUDA toolkit
• generate 3D point clouds and sky masks from images

When to choose

• you need fast, dependency-free depth estimation on CPU or edge devices
• you want to embed monocular depth inference in native applications without Python
• you need bit-exact parity with Depth Anything 3 outputs
• you want quantized models (down to ~99 MB) for lightweight deployment

When to avoid

• you need training or fine-tuning of depth models, not just inference
• you prefer a Python/PyTorch ecosystem with easy experimentation
• you need multi-view stereo reconstruction beyond what DA3 provides

Facets

library · maturity active

machine-learning computer-vision image-processing llm-inference computer-vision machine-learning deep-learning graphics cpp windows cli depth-estimation ggml gguf monocular-depth camera-pose 3d-reconstruction point-cloud depth-anything cpu-inference c-api linux macos gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="localai-org/depth-anything.cpp")