dcharatan/pixelsplat

Health v2 · maintenance only

Flags: no_releases

Adoption not part of the score

1274 stars · 79 forks observed · 2026-08-28

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

pixelSplat is a PyTorch implementation of a feed-forward model that reconstructs 3D radiance fields parameterized by 3D Gaussian primitives from pairs of images in a single forward pass. It enables real-time, memory-efficient novel view synthesis and fast 3D reconstruction, benchmarked on RealEstate10k and ACID datasets.

Use cases

• reconstruct 3D scenes from image pairs
• novel view synthesis from sparse views
• generate 3D Gaussian splats from images
• render depth maps from reconstructed Gaussians
• research on generalizable 3D reconstruction
• train models on RealEstate10k or ACID datasets

When to choose

• you need fast feed-forward 3D reconstruction from two or more images
• you want real-time rendering with 3D Gaussian splatting
• you are doing research on generalizable novel view synthesis
• you need an interpretable, editable 3D radiance field output

When to avoid

• you need per-scene optimization with maximum quality rather than speed
• you lack a CUDA-capable GPU with sufficient memory
• you need reconstruction from a single image
• you want a production-ready application rather than research code

Facets

library · maturity stable

machine-learning deep-learning image-processing graphics computer-vision computer-vision deep-learning machine-learning graphics python 3d-reconstruction gaussian-splatting novel-view-synthesis feed-forward-model cvpr-2024 research-code pytorch linux gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="dcharatan/pixelsplat")