Uni3D: Exploring Unified 3D Representation at Scale

10 Oct 2023  ยท  Junsheng Zhou, Jinsheng Wang, Baorui Ma, Yu-Shen Liu, Tiejun Huang, Xinlong Wang ยท

Scaling up representations for images or text has been extensively investigated in the past few years and has led to revolutions in learning vision and language. However, scalable representation for 3D objects and scenes is relatively unexplored. In this work, we present Uni3D, a 3D foundation model to explore the unified 3D representation at scale. Uni3D uses a 2D initialized ViT end-to-end pretrained to align the 3D point cloud features with the image-text aligned features. Via the simple architecture and pretext task, Uni3D can leverage abundant 2D pretrained models as initialization and image-text aligned models as the target, unlocking the great potential of 2D models and scaling-up strategies to the 3D world. We efficiently scale up Uni3D to one billion parameters, and set new records on a broad range of 3D tasks, such as zero-shot classification, few-shot classification, open-world understanding and part segmentation. We show that the strong Uni3D representation also enables applications such as 3D painting and retrieval in the wild. We believe that Uni3D provides a new direction for exploring both scaling up and efficiency of the representation in 3D domain.

PDF Abstract

Results from the Paper


 Ranked #1 on Zero-shot 3D classification on Objaverse LVIS (using extra training data)

     Get a GitHub badge
Task Dataset Model Metric Name Metric Value Global Rank Uses Extra
Training Data
Result Benchmark
Zero-Shot Transfer 3D Point Cloud Classification ModelNet40 Uni3D Accuracy (%) 88.2 # 1
Zero-shot 3D classification Objaverse LVIS Uni3D Top 1 Accuracy 55.3 # 1
Zero-shot 3D Point Cloud Classification ScanNetV2 Uni3D Top 1 Accuracy % 45.8 # 2
Zero-Shot Transfer 3D Point Cloud Classification ScanObjectNN Uni3D OBJ_ONLY Accuracy(%) 65.3 # 2

Methods