zju3dv/InfiniDepth

Health v2 · maintenance only

Adoption not part of the score

1073 stars · 47 forks observed · 2026-08-28

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

InfiniDepth is a CVPR 2026 research library for monocular depth estimation that represents depth as neural implicit fields, allowing depth queries at continuous 2D coordinates for arbitrary-resolution and fine-grained outputs. It also supports metric depth estimation with depth sensor input and single-image novel view synthesis via 3D Gaussian Splatting.

Use cases

• estimate depth from a single RGB image at any output resolution
• generate 16K-resolution depth maps from photos
• recover fine-grained point clouds from one image
• estimate metric depth using RGB plus a depth sensor
• create 3D Gaussian Splatting scenes from a single view
• perform novel view synthesis with large viewpoint shifts
• evaluate depth models on high-resolution 4K benchmarks

When to choose

• you need depth maps at resolutions beyond the input image grid
• you want fine geometric detail recovery from single images
• you need metric depth from RGB plus depth sensor data
• you want single-image 3DGS reconstruction or novel view synthesis

When to avoid

• you need real-time depth on edge devices
• you want a simple plug-and-play depth estimator without GPU research setup
• you need multi-view or stereo depth estimation

Facets

library · maturity active

machine-learning deep-learning computer-vision image-processing graphics computer-vision machine-learning deep-learning graphics artificial-intelligence python cross-platform depth-estimation monocular-depth neural-implicit-fields 3d-reconstruction gaussian-splatting novel-view-synthesis metric-depth cvpr-2026 gradio-demo research-code linux gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="zju3dv/InfiniDepth")