Objaverse-XL: A Universe of 10M+ 3D Objects

By Matt Deitke, Ruoshi Liu, Matthew Wallingford, Huong Ngo, Oscar Michel, Aditya Kusupati, Alan Fan, Christian Laforte, Vikram S. Voleti, S. Gadre and 7 more. Neural Information Processing Systems, 2023

873 citations recorded by the source corpus.

Natural language processing and 2D vision models have attained remarkable proficiency on many tasks primarily by escalating the scale of training data. However, 3D vision tasks have not seen the same progress, in part due to the challenges of acquiring high-quality 3D data. In this work, we present Objaverse-XL, a dataset of over 10 million 3D objects. Our dataset comprises deduplicated 3D objects from a diverse set of sources, including manually designed objects, photogrammetry scans of landmarks and everyday items, and professional scans of historic and antique artifacts. Representing the largest scale and diversity in the realm of 3D datasets, Objaverse-XL enables significant new possibilities for 3D vision. Our experiments demonstrate the improvements enabled with the scale provided by Objaverse-XL. We show that by training Zero123 on novel view synthesis, utilizing over 100 million multi-view rendered images, we achieve strong zero-shot generalization abilities. We hope that releasing Objaverse-XL will enable further innovations in the field of 3D vision at scale.

View Paper: https://www.semanticscholar.org/paper/1b90e9e9734bed6b379ae87d688cb3b887baf597

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Citations in the corpus

Zero123++: a Single Image to Consistent Multi-view Diffusion Base Model (2023, methodology): “We evaluate the LPIPS score [26] of different models on the validation split (subset of Objaverse) to quantitatively compare Zero-1-to-3 [12], Zero-1-to-3 XL [3,12] and Zero123++. SyncDreamer [13] is excluded because it does not support changing the elevation.”
One-2-3-45++: Fast Single Image to 3D Objects with Consistent Multi-View Generation and 3D Diffusion (2023, methodology): “Replacing our consistent multi-view generation module with Zero123XL [10] led to a noticeable performance decline.”
Hunyuan3D 2.0: Scaling Diffusion Models for High Resolution Textured 3D Assets Generation (2025, unknown): “Due to the viewpoint bias and multi-view inconsistency inherent in training-free image diffusion models, multi-view image diffusion was developed to alleviate these issues by utilizing large-scale 3D datasets, such as objaverse and objaverse-xl [20, 19].”
Splatter Image: Ultra-Fast Single-View 3D Reconstruction (2023, background): “While they are not as fast to render as explicit representations, they are fast enough to be effectively used for single-view reconstruction [1, 7].”
Structured 3D Latents for Scalable and Versatile 3D Generation (2024, unknown): “Objaverse-XL [8].”
DROID: A Large-Scale In-The-Wild Robot Manipulation Dataset (2024, unknown): “Examples include ImageNet [11], Kitti [18], Ego4D [19] and LAION [46] in computer vision, Common Crawl [43] and The Pile [17] in natural language processing, and ShapeNet [5] and Objaverse [9, 10] in 3D modeling.”
DL3DV-10K: A Large-Scale Scene Dataset for Deep Learning-based 3D Vision (2023, unknown): “NVS benchmarks are generally split into synthetic benchmarks like the NeRF-synthetic (Blender) [19], ShapeNet [7] and Objaverse [12], featuring 3D CAD models with varied textures and complex geometries but lacking real-world hints such as noise and non-Lambertian effects.”
LGM: Large Multi-View Gaussian Model for High-Resolution 3D Content Creation (2024, methodology): “In contrast to SDS-based optimization methods, feed-forward 3D native methods are able to generate 3D assets within seconds after training on large-scale 3D datasets [11, 12].”
Recent advances in 3D Gaussian splatting (2024, unknown): “Then it trains a U-Net-based network to generate 3DGS from the multi-view images.”
InstantMesh: Efficient 3D Mesh Generation from a Single Image with Sparse-view Large Reconstruction Models (2024, background): “The availability of large-scale 3D datasets [8, 9] enables training highly generalizable reconstruction models for feed-forward image-to-3D creation.”
SV3D: Novel Multi-view Synthesis and 3D Generation from a Single Image using Latent Video Diffusion (2024, methodology): “Follow-up works such as Zero123XL [8] and Stable Zero123 [48] advance the quality of diffusion-based NVS, as well as the trained NeRFs using SDS loss.”
SyncDreamer: Generating Multiview-consistent Images from a Single-view Image (2023, methodology): “To further increase the quality, we may need to use a larger object dataset like Objaverse-XL [12] and manually clean the dataset to exclude some uncommon shapes like complex scene representation, textureless 3D models, and point clouds.”
Text-to-3D using Gaussian Splatting (2023, background): “Another line of work lies in training or fine-tuning diffusion models directly on 3D datasets (e.g. ShapeNet [9] and Objaverse [18, 19]) to achieve more consistent results with advanced guidance [11, 16, 24, 30, 33, 35, 37, 63, 64, 75, 82].”
DreamGaussian: Generative Gaussian Splatting for Efficient 3D Content Creation (2023, methodology): “A dataset of 30 images collected from previous works (Melas-Kyriazi et al., 2023; Liu et al., 2023a; Tang et al., 2023b; Liu et al., 2023c) and Internet covering various objects is used.”
Instant3D: Fast Text-to-3D with Sparse-View Generation and Large Reconstruction Model (2023, methodology): “While Laion5B has 5 billion text-image pairs, Objaverse-XL (Deitke et al., 2023a), the largest public 3D dataset, contains only 10 million 3D assets with less diversity and poorer annotations.”
Stable Video Diffusion: Scaling Latent Video Diffusion Models to Large Datasets (2023, methodology): “Finally, we demonstrate that our video-pretraining can serve as a strong implicit 3D prior, by tuning our image-to-video models on multi-view generation in Section 4.5 and outperform concurrent work, in particular Zero123XL [14, 57] and Sync-Dreamer [58] in terms of multi-view consistency.”