Geometric Foundation Model Distillation for Efficient Lunar 3D Reconstruction

• URL: http://arxiv.org/abs/2607.01851v1
• Date: Thu, 02 Jul 2026 08:11:32 GMT
• Title: Geometric Foundation Model Distillation for Efficient Lunar 3D Reconstruction
• Abstract summary: Large 3D foundation models such as MASt3R achieve state-of-the-art stereo reconstruction but are computationally demanding for deployment under strict hardware constraints.<n>We study how far such models can be compressed through knowledge distillation, using lunar stereo reconstruction as a challenging and practically relevant case study.
• Score: 0.13999481573773068
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Large 3D foundation models such as MASt3R achieve state-of-the-art stereo reconstruction but are computationally demanding for deployment under strict hardware constraints -- a critical limitation in domains such as planetary exploration, where onboard computing is severely restricted. We study how far such models can be compressed through knowledge distillation, using lunar stereo reconstruction as a challenging and practically relevant case study. Starting from a 688M-parameter MASt3R teacher fine-tuned on lunar imagery, we distill its dense geometric predictions into a family of lightweight students spanning different encoder types (CNN vs ViT), decoder widths and depths, and training strategies. To bridge the dimensional mismatch between teacher and student, we propose a structured SVD-based initialization that projects the teacher's decoder weights into the student's smaller latent space, yielding a warm start that significantly improves convergence and final performance. Based on our results on lunar data, we can obtain a distilled student that retains most of teacher's reconstruction accuracy while reducing the model size up to 7 times, and even outperforms a baseline trained directly with sparse ground-truth annotations. Beyond compression, our study highlights both principles and practical insights for distilling geometric foundation models: a convolutional encoder underperforms transformer-based alternatives (though pretraining availability remains a confounding factor), preserving encoder capacity is more critical than maintaining a large decoder, feature-level distillation consistently outperforms output-only supervision, and SVD-based initialization improves optimisation stability. These findings provide practical guidelines for deploying 3D reconstruction models in resource-constrained environments.

Related papers

• Reliev3R: Relieving Feed-forward Reconstruction from Multi-View Geometric Annotations [98.66466590444553]
  We propose Reliev3R, a weakly-supervised paradigm for training FFRMs from scratch without cost-prohibitive multi-view geometric annotations.<n>At the core of Reliev3R, we design an ambiguity-aware relative depth loss and a trigonometry-based reprojection loss to facilitate supervision for multi-view geometric consistency.
  arXiv  Detail & Related papers  (2026-04-01T06:46:54Z)
• S-MUSt3R: Sliding Multi-view 3D Reconstruction [17.018626984951823]
  This work proposes S-MUSt3R, a simple and efficient pipeline that extends the limits of foundation models for monocular 3D reconstruction.<n>We show that S-MUSt3R runs successfully on long RGB sequences and produces accurate and consistent 3D reconstruction.
  arXiv  Detail & Related papers  (2026-02-04T13:07:14Z)
• Distill3R: A Pipeline for Democratizing 3D Foundation Models on Commodity Hardware [2.676349883103404]
  Distill3R is a framework designed to distill the geometric reasoning of 3D foundation models into compact students fully trainable on a single workstation.<n>Our methodology centers on two primary innovations: (1) an offline caching pipeline that decouples heavy teacher inference from the training loop through compressed supervision signals, and (2) a confidence-aware distillation loss that leverages teacher uncertainty to enable training on commodity hardware.
  arXiv  Detail & Related papers  (2026-01-31T18:56:45Z)
• TAP-CT: 3D Task-Agnostic Pretraining of Computed Tomography Foundation Models [39.00742360251856]
  Existing foundation models (FMs) in the medical domain often require extensive fine-tuning or rely on training resource-intensive decoders.<n>We introduce a suite of task-agnostic pretraining of CT foundation models (TAP-CT)<n>Our approach incorporates targeted modifications to patch embeddings, positional encodings, and volumetric augmentations, making the architecture depth-aware.
  arXiv  Detail & Related papers  (2025-11-30T12:43:15Z)
• Foundry: Distilling 3D Foundation Models for the Edge [15.177160184417575]
  Foundation Model Distillation (FMD) is a new paradigm for compressing large SSL models into compact proxies.<n>We present Foundry, the first implementation of FMD for 3D point clouds.
  arXiv  Detail & Related papers  (2025-11-25T07:53:56Z)
• MoRE: 3D Visual Geometry Reconstruction Meets Mixture-of-Experts [50.37005070020306]
  MoRE is a dense 3D visual foundation model based on a Mixture-of-Experts (MoE) architecture.<n>MoRE incorporates a confidence-based depth refinement module that stabilizes and refines geometric estimation.<n>It integrates dense semantic features with globally aligned 3D backbone representations for high-fidelity surface normal prediction.
  arXiv  Detail & Related papers  (2025-10-31T06:54:27Z)
• On Geometry-Enhanced Parameter-Efficient Fine-Tuning for 3D Scene Segmentation [52.96632954620623]
  We introduce a novel geometry-aware PEFT module specifically designed for 3D point cloud transformers.<n>Our approach sets a new benchmark for efficient, scalable, and geometry-aware fine-tuning of large-scale 3D point cloud models.
  arXiv  Detail & Related papers  (2025-05-28T15:08:36Z)
• FILP-3D: Enhancing 3D Few-shot Class-incremental Learning with Pre-trained Vision-Language Models [59.13757801286343]
  Few-shot class-incremental learning aims to mitigate the catastrophic forgetting issue when a model is incrementally trained on limited data.<n>We introduce the FILP-3D framework with two novel components: the Redundant Feature Eliminator (RFE) for feature space misalignment and the Spatial Noise Compensator (SNC) for significant noise.
  arXiv  Detail & Related papers  (2023-12-28T14:52:07Z)
• Asymmetric Masked Distillation for Pre-Training Small Foundation Models [52.56257450614992]
  Self-supervised foundation models have shown great potential in computer vision thanks to the pre-training paradigm of masked autoencoding. This paper focuses on pre-training relatively small vision transformer models that could be efficiently adapted to downstream tasks. We propose a new asymmetric masked distillation (AMD) framework for pre-training relatively small models with autoencoding.
  arXiv  Detail & Related papers  (2023-11-06T14:44:34Z)
• Robust Geometry-Preserving Depth Estimation Using Differentiable Rendering [93.94371335579321]
  We propose a learning framework that trains models to predict geometry-preserving depth without requiring extra data or annotations. Comprehensive experiments underscore our framework's superior generalization capabilities. Our innovative loss functions empower the model to autonomously recover domain-specific scale-and-shift coefficients.
  arXiv  Detail & Related papers  (2023-09-18T12:36:39Z)
• Towards Efficient Task-Driven Model Reprogramming with Foundation Models [52.411508216448716]
  Vision foundation models exhibit impressive power, benefiting from the extremely large model capacity and broad training data. However, in practice, downstream scenarios may only support a small model due to the limited computational resources or efficiency considerations. This brings a critical challenge for the real-world application of foundation models: one has to transfer the knowledge of a foundation model to the downstream task.
  arXiv  Detail & Related papers  (2023-04-05T07:28:33Z)
• Atlas Based Representation and Metric Learning on Manifolds [0.0]
  We explore the use of a topological manifold, represented as a collection of charts, as the target space of neural network based representation learning tasks. This is achieved by a simple adjustment to the output of an encoder's network architecture plus the addition of a maximal mean discrepancy (MMD) based loss function for regularization.
  arXiv  Detail & Related papers  (2021-06-13T18:05:46Z)

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.