# cvg/depthsplat

[CVPR'25] DepthSplat: Connecting Gaussian Splatting and Depth

Repository: https://github.com/cvg/depthsplat
Canonical: https://ross.abutalabs.com/products/depthsplat
Homepage: https://haofeixu.github.io/depthsplat/
Language: Python
License: MIT
License Family: permissive
Topics: feed-forward-gaussian-splatting, monocular-depth, multi-view-stereo, view-synthesis
Last push: 2026-04-01T12:53:56+00:00

## Health v2 (maintenance only)
Score: 56/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 75, release rhythm 35, longevity 48
- inputs: {"age_days": 685, "days_push": 154, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1242, forks 77 (observed 2026-08-28T04:04:06.449544+00:00)

## What it is
DepthSplat is a PyTorch research library implementing a CVPR 2025 model that connects Gaussian splatting with single/multi-view depth estimation. It provides pre-trained models for feed-forward novel view synthesis and scale-consistent depth prediction from multiple input views.

## Use cases
- render novel views of unseen scenes from a few input images
- estimate multi-view depth from calibrated camera images
- pre-train depth models with Gaussian splatting on unlabeled data
- reconstruct 3D scenes as Gaussians in a feed-forward pass
- benchmark depth estimation on ScanNet, RealEstate10K, and DL3DV
- generate 3D Gaussian reconstructions without per-scene optimization

## When to choose
- you need state-of-the-art feed-forward Gaussian splatting with depth-aware geometry
- you want fast novel view synthesis from sparse views without per-scene training
- you need scale-consistent multi-view depth predictions
- you are researching cross-task interactions between depth estimation and 3D reconstruction

## When to avoid
- you need real-time rendering on consumer hardware without a GPU
- you want classical per-scene Gaussian splatting optimization like vanilla 3DGS
- you need a production-ready application rather than a research codebase
- you lack CUDA-capable hardware or familiarity with PyTorch research workflows

## Facets
- artifact type: library
- maturity: active
- function: machine-learning, deep-learning, image-processing, graphics, simulation
- domain: computer-vision, graphics, machine-learning, deep-learning
- platform: python
- tags: gaussian-splatting, depth-estimation, novel-view-synthesis, multi-view-stereo, monocular-depth, feed-forward, cvpr-2025, research, gpu, linux

## Member repositories
- cvg/depthsplat (main) score 56

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:04:06.449544+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T05:08:24.104827+00:00, confidence not recorded.
  - readme: https://github.com/cvg/depthsplat (fetched 2026-08-28T04:04:06.449544+00:00, sha 1530a19d3ef0)
  - homepage: https://haofeixu.github.io/depthsplat/ (fetched 2026-08-29T12:20:22.042313+00:00, sha bb0461ae56c7)
- Data as of 2026-08-30T08:39:29.467469+00:00.