Research map / 3D vision / 3D generation
TreeSBA: Tree-Transformer for Self-Supervised Sequential Brick Assembly
Mengqi Guo, Chen Li, Yuyang Zhao, Gim Hee Lee
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 |
|---|---|---|---|
| add synthetic pre-training with step-wise action supervision before self-training (ablation) | TreeSBA · ModelNet3-C | mIoU: 26.7 -> 41.6 | Improved |
| add global geometry embedding to brick query (ablation) | TreeSBA · ModelNet3-C | mIoU: 32.1 -> 41.6 | Improved |
| add tree structure embedding to brick query (ablation) | TreeSBA · ModelNet3-C | mIoU: 35.4 -> 41.6 | Improved |
| add action reordering (change root brick) (ablation) | TreeSBA · ModelNet3-C | mIoU: 34.3 -> 41.6 | Improved |
| add silhouette self-training on target data (ablation) | RAD pre-trained model · ModelNet40-C | mIoU: 36.5 -> 45.4 | Improved |
| self-train on 3 classes instead of all 40 classes (ablation) | TreeSBA · ModelNet40-C | mIoU: 45.4 -> 45.1 | No clear effect |
| use a smaller brick (1,2) and more bricks (about 500) (ablation) | TreeSBA · ModelNet3-C | mIoU: 42.7 -> 54.7 | Improved |
| raise voxelization masking threshold beyond 0.4 (ablation) | TreeSBA self-training · MNIST-C, ModelNet3-C, ModelNet40-C | mIoU: drops of 9.9%, 3.0%, and 3.4% | Hurt |
| use class-agnostic joint fine-tuning instead of class-specific models (ablation) | TreeSBA · MNIST-C | mIoU: 64.2 -> 63.9 | No clear effect |
| tree-transformer with parallel decoding versus supervised-learning baseline | TreeSBA · RAD-1k | IoU: 23.0 -> 71.2 | Improved |
Closest papers by meaning
- BrickAnything: Geometry-Conditioned Buildable Brick Generation with Structure-Aware Tokenization (NeurIPS 2026 · similarity 0.86)
- Neural Assembler: Learning to Generate Fine-Grained Robotic Assembly Instructions from Multi-View Images (AAAI 2025 · similarity 0.84)
- IKEA Manuals at Work: 4D Grounding of Assembly Instructions on Internet Videos (NeurIPS 2024 · similarity 0.84)
- PHFormer: Multi-Fragment Assembly Using Proxy-Level Hybrid Transformer (AAAI 2024 · similarity 0.83)
- Imagine: Image-Guided 3D Part Assembly with Structure Knowledge Graph (AAAI 2025 · similarity 0.83)
- Learning Stackable and Skippable LEGO Bricks for Efficient, Reconfigurable, and Variable-Resolution Diffusion Modeling (ICLR 2024 · similarity 0.82)
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.
