Research map / 3D vision / Face and shape
DeFormer: Integrating Transformers with Deformable Models for 3D Shape Abstraction from a Single Image
Di Liu, Xiang Yu, Meng Ye, Qilong Zhangli, Zhuowei Li, Zhixing Zhang, Dimitris N. Metaxas
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 |
|---|---|---|---|
| use DeFormer (bi-channel Transformer with deformable superquadrics, 4 primitives) over Neural Parts (5 primitives) | single-image shape abstraction · ShapeNet 13 categories | average IoU: 0.614 -> 0.634 | Improved |
| use DeFormer instead of Neural Parts | single-image shape abstraction · ShapeNet 13 categories | average Chamfer-L1: 0.114 -> 0.089 | Improved |
| add external model loss L_ext (ablation) | DeFormer · ShapeNet airplane | IoU: 0.633 -> 0.641 | Improved |
| add generalized model loss L_gen (ablation) | DeFormer · ShapeNet car | IoU: 0.723 -> 0.729 | Improved |
| add both cycle-consistency losses L_gcc and L_icc (ablation) | DeFormer · ShapeNet car | IoU: 0.721 -> 0.729 | Improved |
| add image model loss L_sigma (ablation) | DeFormer · ShapeNet airplane | IoU: 0.648 -> 0.641 | Hurt |
| add L_icc image-level cycle loss (ablation) | DeFormer · ShapeNet airplane | IoU: 0.652 -> 0.641 | Hurt |
| increase number of primitives (ablation) | DeFormer · ShapeNet | IoU | No clear effect |
| use different random seeds for primitive initialization (ablation) | DeFormer · ShapeNet airplane | semantic part consistency | No clear effect |
Later papers that compared with this method
8 later accepted papers report Deformer as a baseline in their results tables; 7 of them report a better number on most of the results they share with it.
- DuRP: Dual-Stage Physics-Embedded Learning for Joint Radiance and Polarization Restoration (ICML 2026 · better on 6 of 6 shared results)
- Pose-Guided Temporal Enhancement for Robust Low-Resolution Hand Reconstruction (CVPR 2025 · better on 0 of 1 shared result)
- ExtPose: Robust and Coherent Pose Estimation by Extending ViTs (ICML 2025 · better on 1 of 1 shared result)
- Prior-aware Dynamic Temporal Modeling Framework for Sequential 3D Hand Pose Estimation (ICCV 2025 · better on 1 of 1 shared result)
- MaskHand: Generative Masked Modeling for Robust Hand Mesh Reconstruction in the Wild (ICCV 2025 · better on 1 of 1 shared result)
- MS-MANO: Enabling Hand Pose Tracking with Biomechanical Constraints (CVPR 2024 · better on 3 of 3 shared results)
- A Simple Baseline for Efficient Hand Mesh Reconstruction (CVPR 2024 · better on 2 of 2 shared results)
- ScoreHypo: Probabilistic Human Mesh Estimation with Hypothesis Scoring (CVPR 2024 · better on 2 of 2 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
- Canonical 3D Deformer Maps: Unifying parametric and non-parametric methods for dense weakly-supervised category reconstruction (NeurIPS 2020 · similarity 0.88)
- Unsupervised Learning of Dense Shape Correspondence (CVPR 2019 · similarity 0.88)
- Learning Category-Specific Deformable 3D Models for Object Reconstruction (TPAMI 2016 · similarity 0.88)
- Extracting Deformation-Aware Local Features by Learning to Deform (NeurIPS 2021 · similarity 0.87)
- Deformable Mesh Transformer for 3D Human Mesh Recovery (CVPR 2023 · similarity 0.87)
- PyTorchGeoNodes: Enabling Differentiable Shape Programs for 3D Shape Reconstruction (CVPR 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.
