20 datasets found

The Mip-NeRF 360 and Tanks & Temples datasets are used to evaluate the performance of the Pixel-GS method.

Real-world scenes with reflections, including forward-facing scenes with strong reflection effects caused by glass and mirrors.

A dataset for training and testing semantically consistent few-shot view synthesis models

A digital replica of indoor spaces for training and testing NeRF-based models

A dataset of images of buildings and landmarks, used for training and testing image-based rendering and view synthesis algorithms.

The dataset used in the paper is the LLFF dataset, which contains real-world scenes and is used for training and testing the proposed neural radiance field model.

LLFF Real dataset comprises 24 scenes captured from handheld cellphones with 20 − 30 images each.

Real-world forward-facing dataset for view synthesis, containing 1008x756 images.

A dataset of synthetic photorealistic images for training and testing material-aware diffusion models

The dataset used in the paper for training and testing the NeRF model.

Soft 3D Reconstruction for View Synthesis

Nex: Real-Time View Synthesis with Neural Basis Expansion

Stereo Magnification: Learning View Synthesis using Multi-Plane Images

NeRF: Representing Scenes as Neural Radiance Fields for View Synthesis

Real-world multi-view datasets DTU and LLFF for view synthesis from sparse inputs.

MobileNeRF: Exploiting the Polygon Rasterization Pipeline for Efficient Neural Field Rendering on Mobile Architectures

Recent work has addressed this issue by “baking” NeRFs into a sparse 3D voxel grid [21, 51]. For example, Hedman et al. introduced Sparse Neural Radiance Grids (SNeRG) [21],...

NeRF [33] has demonstrated amazing ability to synthesize images of 3D scenes from novel views. However, they rely upon specialized volumetric rendering algorithms based on ray...

The DTU dataset is a large-scale dataset for 3D reconstruction and editing. It contains 15 scenes with images of 1600x1200 resolution and accompanying foreground masks.

The LLFF dataset contains 8 forward-facing scenes. Following [19, 26, 35, 42], we take every 8-th image as the novel views for testing. The input views are evenly sampled across...