LLM2R: From Large Language Models to 3D Reassemblers

Hugging Face Daily Papers Papers
3d-assembly large-language-models point-cloud geometric-tokens lora computer-vision

Summary

LLM2R casts 3D part assembly as sequence modeling with a pretrained decoder-only LLM, using geometric tokens and learnable summary tokens to predict all part poses in a single non-autoregressive pass, achieving state-of-the-art accuracy on PartNet, PartNeXt, and Breaking Bad benchmarks.

3D part assembly aims to reconstruct a complete shape from a set of unassembled parts by predicting a rigid SE(3) transformation for each part. Because the placement of a part is rarely determined by its own geometry, the task demands reasoning over the entire part set, and existing methods, which rely on pairwise relations or iterative generation and verification, scale poorly to objects with many parts. We present LLM2R, which casts part assembly as sequence modeling with a pretrained decoder-only large language model (LLM): a Point-BERT encoder and a Q-Former compress each part into a block of geometric tokens, the blocks of all parts are concatenated, and a causal LLM adapted with Low-Rank Adaptation (LoRA) processes the sequence in one forward pass. Causal attention, however, lets the hidden states of a part observe only the parts that precede it. LLM2R therefore appends a block of learnable summary tokens after the last part: these tokens follow every part, so their states aggregate the whole assembly at a constant token cost, and a part-aware pose head fuses each part's states with them before predicting its rotation and translation. One architecture thus handles semantic and geometric assembly and predicts all poses in a single non-autoregressive pass. LLM2R achieves the highest part accuracy and success rate on all three PartNet categories, the best rotation error, translation error, and part accuracy on PartNeXt, and the lowest translation error and fastest inference on Breaking Bad, where only a fracture-specialized method is more accurate. Its advantage grows with the number of parts: at 100 parts it exceeds GARF by 33.5 points in part accuracy on Breaking Bad and Assembler by 29.1 points on PartNet, and it transfers zero-shot to unseen categories with more than 16 points higher part accuracy than the strongest of three baselines.
Original Article
View Cached Full Text

Cached at: 10/21/26, 10:20 AM

Paper page - LLM2R: From Large Language Models to 3D Reassemblers

Source: https://huggingface.co/papers/2610.18604

Abstract

LLM2R treats 3D part assembly as sequence modeling with a pretrained decoder-only LLM, using geometric tokens and learnable summary tokens to predict all part poses in a single pass with improved scalability and accuracy.

3D part assembly aims to reconstruct a complete shape from a set of unassembled parts by predicting a rigidSE(3) transformationfor each part. Because the placement of a part is rarely determined by its own geometry, the task demands reasoning over the entire part set, and existing methods, which rely on pairwise relations or iterative generation and verification, scale poorly to objects with many parts. We present LLM2R, which casts part assembly as sequence modeling with a pretraineddecoder-only large language model(LLM): aPoint-BERTencoder and aQ-Formercompress each part into a block ofgeometric tokens, the blocks of all parts are concatenated, and a causal LLM adapted with Low-Rank Adaptation (LoRA) processes the sequence in one forward pass.Causal attention, however, lets the hidden states of a part observe only the parts that precede it. LLM2R therefore appends a block oflearnable summary tokensafter the last part: these tokens follow every part, so their states aggregate the whole assembly at a constant token cost, and apart-aware pose headfuses each part’s states with them before predicting its rotation and translation. One architecture thus handles semantic and geometric assembly and predicts all poses in a singlenon-autoregressivepass. LLM2R achieves the highest part accuracy and success rate on all three PartNet categories, the best rotation error, translation error, and part accuracy on PartNeXt, and the lowest translation error and fastest inference on Breaking Bad, where only a fracture-specialized method is more accurate. Its advantage grows with the number of parts: at 100 parts it exceeds GARF by 33.5 points in part accuracy on Breaking Bad and Assembler by 29.1 points on PartNet, and it transfers zero-shot to unseen categories with more than 16 points higher part accuracy than the strongest of three baselines.

View arXiv pageView PDFAdd to collection

Get this paper in your agent:

hf papers read 2610\.18604

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2610.18604 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2610.18604 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2610.18604 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

Imagine3D-LLM: Teaching MLLMs to Imagine 3D Scenes Before Answering

Hugging Face Daily Papers

Imagine3D-LLM proposes teaching multimodal LLMs to imagine a compact 3D representation of a scene (decoded from learnable summary tokens as 3D Gaussian Splatting via photometric reconstruction loss) before answering multi-view spatial reasoning questions, outperforming prior 3D-aware MLLM approaches on multiple benchmarks.

2.5-D Decomposition for LLM-Based Spatial Construction

arXiv cs.AI

This paper introduces a neuro-symbolic pipeline using 2.5-D decomposition to improve LLM-based spatial construction accuracy by offloading vertical coordinate calculation to a deterministic executor, achieving high accuracy on benchmarks and edge hardware.