dxgl/objaverse-1k

DX.GL Objaverse-1K: Dense Multi-View Datasets for 3D Vision

1026 objects × 196 views × 1024x1024 resolution × 6 modalities — ready for nerfstudio out of the box.

Dense multi-view datasets rendered from curated Objaverse 3D models via DX.GL. Each object includes calibrated camera poses, depth maps (8-bit + 16-bit), normal maps, binary masks, and point clouds — suitable for novel view synthesis, 3D reconstruction, monocular depth/normal estimation, segmentation, and more.

Quick Start

# Download a single dataset
pip install requests
python download_all.py --object "chair"

# Train with nerfstudio
pip install nerfstudio
ns-train splatfacto --data ./dxgl-datasets/chair \
  --max-num-iterations 20000 \
  --pipeline.model.sh-degree 3 \
  --pipeline.model.background-color white

Or download everything:

python download_all.py           # all 1026 datasets
python download_all.py --output ./data --no-extract  # keep ZIPs

What's in Each Dataset ZIP

dataset/
├── images/           # RGB frames (PNG, transparent RGBA background)
│   ├── frame_00000.png
│   └── ...
├── depth/            # 8-bit grayscale depth maps
├── depth_16bit/      # 16-bit grayscale depth maps (higher precision)
├── normals/          # World-space normal maps
├── masks/            # Binary alpha masks
├── transforms.json   # Camera poses (nerfstudio / instant-ngp format)
└── points3D.ply      # Sparse point cloud for initialization

Specs

Why This Dataset

Existing Objaverse render datasets target diffusion model training with sparse views (6-38) at low resolution (256-512px). This dataset targets dense 3D vision tasks with more views, higher resolution, and richer modalities:

Use Cases

• —Novel View Synthesis — Train or evaluate NeRF, 3DGS, Zip-NeRF, Instant-NGP with dense calibrated views
• —3D Reconstruction — Multi-view surface reconstruction (NeuS, VolSDF, Neuralangelo)
• —Monocular Depth Estimation — 196×1026 paired RGB+depth images with pixel-perfect ground truth
• —Surface Normal Estimation — Paired RGB+normal maps for training or benchmarking
• —Object Segmentation / Matting — Perfect binary masks for figure-ground separation
• —Single-Image 3D — Training feed-forward models (LRM, InstantMesh, One-2-3-45)
• —Multi-View Consistency Evaluation — Ground-truth views for evaluating generative models (Zero123, MVDream, SV3D)
• —Point Cloud Processing — Calibrated point clouds for completion, classification, and registration

Categories

Objects span 546 LVIS categories. Top categories: racecar (10), chair (10), seashell (8), earphone (7), globe (7), telephone (7), trafficlight (6), motorscooter (6), pickuptruck (6), armoire (6), guitar (6), fire_extinguisher (6), dog (5), handbag (5), shoe (5), helmet (5), watch (5), rifle (5), ring (5), starfish (5).

Camera Distribution

Views are distributed on a full sphere (±89° elevation) using a golden-angle Fibonacci spiral. The distribution is uniform in solid angle — optimized for NeRF/3DGS training.

Rendering Pipeline

Datasets are rendered using DX.GL's cloud GPU rendering pipeline:

• —Lighting: Studio HDRI environment with physically-based rendering (PBR)
• —Camera: Fibonacci golden-angle sphere distribution with calibrated intrinsics
• —Depth: Tight near/far planes derived from model bounding sphere for maximum precision
• —Point cloud: Back-projected from depth maps
• —Background: Neutral; binary masks provided for background removal

License

Objects are licensed under their original Creative Commons terms:

• —8 objects: CC0 1.0 (public domain)
• —1018 objects: CC-BY 4.0 (attribution required)

All source models are from Objaverse / Sketchfab. The rendered datasets (images, depth maps, etc.) inherit the license of their source model.

See the full attribution table below for per-object credits.

Attribution

Citation

@misc{dxgl_objaverse1k_2026,
  title  = {DX.GL Objaverse-1K: Dense Multi-View Datasets for 3D Vision},
  author = {DXGL},
  year   = {2026},
  url    = {https://huggingface.co/datasets/dxgl/objaverse-1k},
  note   = {Dense multi-view datasets with depth, normals, masks, and point clouds for 3D vision research. Rendered via DX.GL.}
}

Links

• —Browse datasets: dx.gl/datasets
• —Pipeline details: dx.gl/for-research
• —API documentation: dx.gl/portal/docs
• —Generate your own: dx.gl/signup

Feedback

Open a Discussion on this repo if you find issues with depth accuracy, mask quality, camera calibration, or view distribution.