Depth Anything 3 — GGUF weights for depth anything.cpp Brought to you by the LocalAI team. GGUF conversions of ByteDance Depth Anything 3, for use with depth anything.cpp — a from scratch C++17 / ggml port. No Python, no PyTorch, no CUDA toolkit at inference: one self contained GGUF file plus a small native library and CLI, faster than PyTorch on CPU and bit exact against the original (correlation 1.0, verified component by component). Given an image, the engine recovers a dense depth map, per pixel confidence , camera extrinsics (3×4) and intrinsics (3×3) , an optional sky mask, a back projected 3D point cloud , and exports to glb / COLMAP / PLY . Files in this repo Each GGUF is fully self contained — every dimension, hyperparameter and preprocessing constant is baked into the file; the loader reads them, nothing is hardcoded. File Source checkpoint Backbone Depth type Output depth anything small f32.gguf DA3 SMALL ViT S relative depth + conf + pose depth anything base f32.gguf DA3 BASE ViT B relative depth + conf + pose depth anything base f16.gguf DA3 BASE ViT B relative depth + conf + pose depth anything base q8 0.gguf DA3 BASE ViT B relative depth + conf + pose (near lossless)…
We use cookies for essential functionality and analytics. You can accept or reject analytics cookies.Cookie policy