MVImgNet: Large-Scale Multi-View Real-World Images for 3D Vision

MVImgNet provides a large-scale collection of real-world multi-view images for 3D vision tasks, published by GAP-LAB-CUHK-SZ at CVPR 2023. The data supports 3D reconstruction, NeRF, and Gaussian Splatting using image sequences of physical objects.

Use Cases

• Novel view synthesis using multi-view image sequences to train Neural Radiance Fields (NeRF)
• 3D reconstruction of real-world objects using multi-view stereo (MVS) techniques
• Point cloud generation from image-based 3D understanding pipelines

Strengths

• CVPR 2023 peer-reviewed origin
• Real-world data capture rather than synthetic generation
• Optimized for 3D Gaussian Splatting and NeRF

Limitations

• High storage requirements typical of multi-view image datasets
• Lack of explicit column or metadata documentation in the provided summary

Provenance

Source

GAP-LAB-CUHK-SZ (CVPR 2023)

Collection Method

Real-world image capture

Freshness

Last updated July 2025; originally presented at CVPR 2023.