ULIP-2: Towards Scalable Multimodal Pre-training for 3D Understanding¶

Recent advancements in multimodal pre-training have shown promising efficacy in 3D representation learning by aligning multimodal features across 3D shapes, their 2D counterparts, and language descriptions. However, the methods used by existing frameworks to curate such multimodal data, in particular language descriptions for 3D shapes, are not scalable, and the collected language descriptions are not diverse. To address this, we introduce ULIP-2, a simple yet effective tri-modal pre-training framework that leverages large multimodal models to automatically generate holistic language descriptions for 3D shapes. It only needs 3D data as input, eliminating the need for any manual 3D annotations, and is therefore scalable to large datasets. ULIP-2 is also equipped with scaled-up backbones for better multimodal representation learning. We conduct experiments on two large-scale 3D datasets, Objaverse and ShapeNet, and augment them with tri-modal datasets of 3D point clouds, images, and language for training ULIP-2. Experiments show that ULIP-2 demonstrates substantial benefits in three downstream tasks: zero-shot 3D classification, standard 3D classification with fine-tuning, and 3D captioning (3D-tolanguage generation). It achieves a new SOTA of 50.6% (top1) on Objaverse-LVIS and 84.7% (top-1) on ModelNet40 in zero-shot classification. In the ScanObjectNN benchmark for standard fine-tuning, ULIP-2 reaches an overall accuracy of 91.5% with a compact model of only 1.4 million parameters. ULIP-2 sheds light on a new paradigm for scalable multimodal 3D representation learning without human annotations and shows significant improvements over existing baselines. The code and datasets are released at https://github.com/salesforce/ULIP.

We present ULIP-2, a novel framework for multimodal 3D representation learning. By leveraging large multimodal models for language description generation and scaling up the multimodal 3D pre-training, ULIP-2 not only addresses the quality and scalability challenges in existing multimodal 3D datasets but also demonstrates significant improvements in all downstream tasks. We also release "ULIP-Objaverse" triplets and "ULIP-ShapeNet" triplets, two large-scale trimodal datasets to foster further research. Limitations. ULIP-2’s pre-training primarily utilizes object-level 3D shape datasets, which inherently differ from scene-level 3D data in their distribution and complexity. Exploring the application of the ULIP-2 framework to scenelevel 3D data understanding, and leveraging the knowledge learned from object-level 3D data for this purpose, represents a compelling avenue for future research. Broader Impact. ULIP-2 aims to minimize human annotation in 3D multimodal pre-training, reducing labor but potentially impacting low-skilled job markets. This dual impact, a common concern in AI advancements, underscores the need for broader considerations in AI research.