3D CoCa: Contrastive Learners are 3D Captioners

TL;DR

3D CoCa tackles 3D captioning from sparse point clouds by unifying contrastive vision-language pre-training with end-to-end 3D caption generation. It employs a frozen CLIP backbone, a 3D scene encoder with point-tokenization and task tokens, a frozen CLIP text encoder, and a multimodal transformer decoder with cross-attention, trained with a joint objective $\mathcal{L}_{Total} = \mathcal{L}_{Con} + \lambda \cdot \mathcal{L}_{Cap}$ to align visual and linguistic representations while producing fluent captions. The model achieves state-of-the-art results on ScanRefer and Nr3D, with substantial CIDEr gains and improved spatial grounding, demonstrating the benefits of integrating rich visual-linguistic priors into 3D captioning without external detectors. This approach offers a scalable, end-to-end paradigm for 3D scene understanding that leverages foundation-model priors and cross-modal alignment to produce more accurate and contextually grounded descriptions.

Abstract

3D captioning, which aims to describe the content of 3D scenes in natural language, remains highly challenging due to the inherent sparsity of point clouds and weak cross-modal alignment in existing methods. To address these challenges, we propose 3D CoCa, a novel unified framework that seamlessly combines contrastive vision-language learning with 3D caption generation in a single architecture. Our approach leverages a frozen CLIP vision-language backbone to provide rich semantic priors, a spatially-aware 3D scene encoder to capture geometric context, and a multi-modal decoder to generate descriptive captions. Unlike prior two-stage methods that rely on explicit object proposals, 3D CoCa jointly optimizes contrastive and captioning objectives in a shared feature space, eliminating the need for external detectors or handcrafted proposals. This joint training paradigm yields stronger spatial reasoning and richer semantic grounding by aligning 3D and textual representations. Extensive experiments on the ScanRefer and Nr3D benchmarks demonstrate that 3D CoCa significantly outperforms current state-of-the-arts by 10.2% and 5.76% in CIDEr at 0.5IoU, respectively. Code will be available at https://github.com/AIGeeksGroup/3DCoCa.

Figures (3)

• Figure 1: Conceptual homepage figure for 3D CoCa, highlighting its architecture (left) and performance (right). Left: The 3D CoCa model unifies contrastive learning and multimodal captioning in one framework. Right:Radar chart comparison of 3D CoCa and previous methods Scan2Cap scan2cap_2021, 3DJCG 3djcg2022, 3D-VLP 3dvlp2024, Vote2Cap-DETR vote2cap2023, Vote2Cap-DETR++ vote2cap++2024 on the ScanRefer chen2020scanrefer benchmark.
• Figure 2: Illustration of a multi-modal Transformer architecture for 3D vision-language understanding. The input point cloud and textual description are processed by CLIP Vision and Text Encoders, respectively. Cross-attention mechanisms fuse these features within a Multi-Modal Decoder, enabling the generation of descriptive captions. The model training is guided by contrastive and captioning losses, promoting effective alignment between visual and textual modalities.
• Figure 3: A visual comparison on the ScanRefer chen2020scanrefer dataset showcasing indoor scenes described by Vote2Cap-DETR++ vote2cap++2024, our method (Ours), and the ground truth (GT), highlighting differences in descriptive accuracy and style.