From Thousands to Billions: 3D Visual Language Grounding via Render-Supervised Distillation from 2D VLMs

• URL: http://arxiv.org/abs/2502.20389v2
• Date: Mon, 09 Jun 2025 14:28:09 GMT
• Title: From Thousands to Billions: 3D Visual Language Grounding via Render-Supervised Distillation from 2D VLMs
• Authors: Ang Cao, Sergio Arnaud, Oleksandr Maksymets, Jianing Yang, Ayush Jain, Sriram Yenamandra, Ada Martin, Vincent-Pierre Berges, Paul McVay, Ruslan Partsey, Aravind Rajeswaran, Franziska Meier, Justin Johnson, Jeong Joon Park, Alexander Sax,
• Abstract summary: LIFT-GS predicts 3D Gaussian representations from point clouds and uses them to render predicted language-conditioned 3D masks into 2D views.<n>LIFT-GS achieves state-of-the-art results with $25.7%$ mAP on open-vocabulary instance segmentation.<n>Remarkably, pretraining effectively multiplies fine-tuning datasets by 2X, demonstrating strong scaling properties.
• Score: 64.28181017898369
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: 3D vision-language grounding faces a fundamental data bottleneck: while 2D models train on billions of images, 3D models have access to only thousands of labeled scenes--a six-order-of-magnitude gap that severely limits performance. We introduce $\textbf{LIFT-GS}$, a practical distillation technique that overcomes this limitation by using differentiable rendering to bridge 3D and 2D supervision. LIFT-GS predicts 3D Gaussian representations from point clouds and uses them to render predicted language-conditioned 3D masks into 2D views, enabling supervision from 2D foundation models (SAM, CLIP, LLaMA) without requiring any 3D annotations. This render-supervised formulation enables end-to-end training of complete encoder-decoder architectures and is inherently model-agnostic. LIFT-GS achieves state-of-the-art results with $25.7\%$ mAP on open-vocabulary instance segmentation (vs. $20.2\%$ prior SOTA) and consistent $10-30\%$ improvements on referential grounding tasks. Remarkably, pretraining effectively multiplies fine-tuning datasets by 2X, demonstrating strong scaling properties that suggest 3D VLG currently operates in a severely data-scarce regime. Project page: https://liftgs.github.io

Related papers

• DINO in the Room: Leveraging 2D Foundation Models for 3D Segmentation [51.43837087865105]
  Vision foundation models (VFMs) trained on large-scale image datasets provide high-quality features that have significantly advanced 2D visual recognition.<n>Their potential in 3D vision remains largely untapped, despite the common availability of 2D images alongside 3D point cloud datasets.<n>We introduce DITR, a simple yet effective approach that extracts 2D foundation model features, projects them to 3D, and finally injects them into a 3D point cloud segmentation model.
  arXiv  Detail & Related papers  (2025-03-24T17:59:11Z)
• Unifying 2D and 3D Vision-Language Understanding [85.84054120018625]
  We introduce UniVLG, a unified architecture for 2D and 3D vision-language learning. UniVLG bridges the gap between existing 2D-centric models and the rich 3D sensory data available in embodied systems.
  arXiv  Detail & Related papers  (2025-03-13T17:56:22Z)
• SplatTalk: 3D VQA with Gaussian Splatting [13.211810095081159]
  Language-guided 3D scene understanding is important for advancing applications in robotics, AR/VR, and human-computer interaction. We introduce SplatTalk, a novel method that uses a generalizable 3D Gaussian Splatting (3DGS) framework to produce 3D tokens suitable for direct input into a pretrained LLM.
  arXiv  Detail & Related papers  (2025-03-08T16:31:48Z)
• A Lesson in Splats: Teacher-Guided Diffusion for 3D Gaussian Splats Generation with 2D Supervision [65.33043028101471]
  We present a novel framework for training 3D image-conditioned diffusion models using only 2D supervision.<n>Most existing 3D generative models rely on full 3D supervision, which is impractical due to the scarcity of large-scale 3D datasets.
  arXiv  Detail & Related papers  (2024-12-01T00:29:57Z)
• ConDense: Consistent 2D/3D Pre-training for Dense and Sparse Features from Multi-View Images [47.682942867405224]
  ConDense is a framework for 3D pre-training utilizing existing 2D networks and large-scale multi-view datasets. We propose a novel 2D-3D joint training scheme to extract co-embedded 2D and 3D features in an end-to-end pipeline.
  arXiv  Detail & Related papers  (2024-08-30T05:57:01Z)
• Improving 2D Feature Representations by 3D-Aware Fine-Tuning [17.01280751430423]
  Current visual foundation models are trained purely on unstructured 2D data. We show that fine-tuning on 3D-aware data improves the quality of emerging semantic features.
  arXiv  Detail & Related papers  (2024-07-29T17:59:21Z)
• Open Vocabulary 3D Scene Understanding via Geometry Guided Self-Distillation [67.36775428466045]
  We propose Geometry Guided Self-Distillation (GGSD) to learn superior 3D representations from 2D pre-trained models. Due to the advantages of 3D representation, the performance of the distilled 3D student model can significantly surpass that of the 2D teacher model.
  arXiv  Detail & Related papers  (2024-07-18T10:13:56Z)
• Cross-Modal Self-Training: Aligning Images and Pointclouds to Learn Classification without Labels [69.55622471172941]
  Large-scale vision 2D vision language models, such as CLIP can be aligned with a 3D encoder to learn generalizable (open-vocabulary) 3D vision models. We propose an optimization framework: Cross-MoST: Cross-Modal Self-Training, to improve the label-free classification performance of a zero-shot 3D vision model.
  arXiv  Detail & Related papers  (2024-04-15T21:30:50Z)
• Weakly Supervised 3D Object Detection via Multi-Level Visual Guidance [72.6809373191638]
  We propose a framework to study how to leverage constraints between 2D and 3D domains without requiring any 3D labels. Specifically, we design a feature-level constraint to align LiDAR and image features based on object-aware regions. Second, the output-level constraint is developed to enforce the overlap between 2D and projected 3D box estimations. Third, the training-level constraint is utilized by producing accurate and consistent 3D pseudo-labels that align with the visual data.
  arXiv  Detail & Related papers  (2023-12-12T18:57:25Z)
• PonderV2: Pave the Way for 3D Foundation Model with A Universal Pre-training Paradigm [114.47216525866435]
  We introduce a novel universal 3D pre-training framework designed to facilitate the acquisition of efficient 3D representation. For the first time, PonderV2 achieves state-of-the-art performance on 11 indoor and outdoor benchmarks, implying its effectiveness.
  arXiv  Detail & Related papers  (2023-10-12T17:59:57Z)
• Uni3D: Exploring Unified 3D Representation at Scale [66.26710717073372]
  We present Uni3D, a 3D foundation model to explore the unified 3D representation at scale. Uni3D uses a 2D ViT end-to-end pretrained to align the 3D point cloud features with the image-text aligned features. We show that the strong Uni3D representation also enables applications such as 3D painting and retrieval in the wild.
  arXiv  Detail & Related papers  (2023-10-10T16:49:21Z)
• Act3D: 3D Feature Field Transformers for Multi-Task Robotic Manipulation [18.964403296437027]
  Act3D represents the robot's workspace using a 3D feature field with adaptive resolutions dependent on the task at hand. It samples 3D point grids in a coarse to fine manner, featurizes them using relative-position attention, and selects where to focus the next round of point sampling.
  arXiv  Detail & Related papers  (2023-06-30T17:34:06Z)
• 3D-to-2D Distillation for Indoor Scene Parsing [78.36781565047656]
  We present a new approach that enables us to leverage 3D features extracted from large-scale 3D data repository to enhance 2D features extracted from RGB images. First, we distill 3D knowledge from a pretrained 3D network to supervise a 2D network to learn simulated 3D features from 2D features during the training. Second, we design a two-stage dimension normalization scheme to calibrate the 2D and 3D features for better integration. Third, we design a semantic-aware adversarial training model to extend our framework for training with unpaired 3D data.
  arXiv  Detail & Related papers  (2021-04-06T02:22:24Z)
• Cycle-Consistent Generative Rendering for 2D-3D Modality Translation [21.962725416347855]
  We learn a module that generates a realistic rendering of a 3D object and infers a realistic 3D shape from an image. By leveraging generative domain translation methods, we are able to define a learning algorithm that requires only weak supervision, with unpaired data. The resulting model is able to perform 3D shape, pose, and texture inference from 2D images, but can also generate novel textured 3D shapes and renders.
  arXiv  Detail & Related papers  (2020-11-16T15:23:03Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.