Towards Large-scale 3D Representation Learning with Multi-dataset Point Prompt Training

TL;DR

The paper tackles the data bottleneck in 3D representation learning by enabling large-scale pre-training across multiple datasets without suffering negative transfer. It introduces Point Prompt Training (PPT), which combines a Domain Prompt Adapter and Prompt-driven Normalization with Language-guided Categorical Alignment to learn a shared backbone that adapts to diverse datasets and unifies label spaces. Through extensive ablations and benchmarks on indoor and outdoor tasks, PPT achieves state-of-the-art performance with a single weight-shared model and proves effective as both supervised and unsupervised pre-training. This approach holds promise for scalable 3D pre-training and cross-domain generalization, with practical impact for 3D scene understanding in varied environments.

Abstract

The rapid advancement of deep learning models often attributes to their ability to leverage massive training data. In contrast, such privilege has not yet fully benefited 3D deep learning, mainly due to the limited availability of large-scale 3D datasets. Merging multiple available data sources and letting them collaboratively train a single model is a potential solution. However, due to the large domain gap between 3D point cloud datasets, such mixed supervision could adversely affect the model's performance and lead to degenerated performance (i.e., negative transfer) compared to single-dataset training. In view of this challenge, we introduce Point Prompt Training (PPT), a novel framework for multi-dataset synergistic learning in the context of 3D representation learning that supports multiple pre-training paradigms. Based on this framework, we propose Prompt-driven Normalization, which adapts the model to different datasets with domain-specific prompts and Language-guided Categorical Alignment that decently unifies the multiple-dataset label spaces by leveraging the relationship between label text. Extensive experiments verify that PPT can overcome the negative transfer associated with synergistic learning and produce generalizable representations. Notably, it achieves state-of-the-art performance on each dataset using a single weight-shared model with supervised multi-dataset training. Moreover, when served as a pre-training framework, it outperforms other pre-training approaches regarding representation quality and attains remarkable state-of-the-art performance across over ten diverse downstream tasks spanning both indoor and outdoor 3D scenarios.

Figures (4)

• Figure 1: Multi-dataset synergistic training with Point Prompt Training (PPT). (a) Our PPT Framework is comprised of two key components: 1. The domain prompt adapter adapts the backbone to various dataset-specific contexts with a set of domain-specific prompts; 2. The categorical alignment process empowers the model to effectively undergo training within multiple category spaces concurrently in the supervised setting. (b) The Result Comparison plot reveals that PPT delivers state-of-the-art performance across both datasets only with one single shared-weight backbone, and fine-tuning on any single specific dataset can further enhance the results.
• Figure 2: Prompt adapter and categorical alignment. (a) As a prompt adapter, Prompt-driven Normalization adaptly encodes domain-specific prompts into the scale and shift vectors in normalization layers. This adaptation helps adapt the model to the specific dataset domain. (b) Language-guided Categorical Alignment aligns point representations to a unified category-language embedding, shared across all datasets and extracted by a pre-trained text encoder.
• Figure 3: Domain prompt adapters.
• Figure 5: Loss curve.