Research map / 3D vision / Face and shape
CRAG: Can 3D Generative Models Help 3D Assembly?
Zeyu Jiang, Sihang Li, Siqi Tan, Chenyang Xu, Juexiao Zhang, Julia Galway-Witham, Xue Wang, Scott A. Williams, Radu Iovita, Chen Feng, Jing Zhang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace PTv3 task-specific encoder with shared TripoSG VAE embedding (ablation) | CRAG assembly branch · PartNeXt (Complete) | Part Accuracy: 58.19 -> 60.01 | Improved |
| add reference image condition to assembly-only model (ablation) | CRAG assembly branch · PartNeXt (Complete) | Part Accuracy: 60.01 -> 62.29 | Improved |
| add generation branch with Joint Adapter (with image) (ablation) | CRAG · PartNeXt (Complete) | Part Accuracy: 62.29 -> 65.89 | Improved |
| add generation branch with Joint Adapter (no image) (ablation) | CRAG · PartNeXt (Complete) | Part Accuracy: 60.01 -> 61.67 | Improved |
| add generation branch with Joint Adapter (no image) (ablation) | CRAG · PartNeXt (Complete) | Chamfer Distance: 3.61 -> 3.31 vs 2.69 with image-only assembly | Hurt |
| add coupled generation to assembly without images | CRAG w/o img vs GARF · PartNeXt (Complete) | Part Accuracy: 58.19 -> 61.67 | Improved |
| add coupled generation under missing parts | CRAG w/o img vs GARF · PartNeXt (Missing parts) | Part Accuracy: 60.71 -> 66.74 | Improved |
| add coupled generation under missing parts | CRAG w/o img vs GARF · PartNeXt (Missing parts) | Chamfer Distance: 5.78 -> 5.17 | Hurt |
| add reference image to CRAG | CRAG · Breaking Bad | Part Accuracy: 94.64 -> 94.68 | No clear effect |
| joint assembly-and-generation instead of image-conditioned Assembler | CRAG · PartNeXt (Complete) | Part Accuracy: 44.18 -> 65.89 | Improved |
| joint formulation versus sequential assembly-then-complete with oracle poses | CRAG w/o img vs MGPC · PartNeXt missing parts | F-Score@5%: 70.54 -> 74.11 | Improved |
| add part-level evidence from assembly branch to image-conditioned generation | TripoSG · PartNeXt generation | F-Score@5%: 73.18 -> 83.87 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- GARF — GARF: Learning Generalizable 3D Reassembly for Real-World Fractures (ICCV 2025) · better on 8 of 8 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Generative 3D Part Assembly via Part-Whole-Hierarchy Message Passing (CVPR 2024 · similarity 0.90)
- Imagine: Image-Guided 3D Part Assembly with Structure Knowledge Graph (AAAI 2025 · similarity 0.88)
- Two by Two: Learning Multi-Task Pairwise Objects Assembly for Generalizable Robot Manipulation (CVPR 2025 · similarity 0.87)
- Neural Assembler: Learning to Generate Fine-Grained Robotic Assembly Instructions from Multi-View Images (AAAI 2025 · similarity 0.87)
- PartGen: Part-level 3D Generation and Reconstruction with Multi-view Diffusion Models (CVPR 2025 · similarity 0.86)
- AutoPartGen: Autoregressive 3D Part Generation and Discovery (NeurIPS 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
