Research map / Graph learning / Graph self-supervised learning
BrickNet: Graph-Backed Generative Brick Assembly
Peter Kulits, Cordelia Schmid
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 graph-backed parametrization instead of pose-based | BrickNet LLM (14b) · BrickNet unconditional, nucleus sampling | steps until connectivity failure: 49.9 -> 96.9 | Improved |
| use graph-backed parametrization instead of pose-based | BrickNet LLM (14b) · BrickNet unconditional, nucleus sampling | steps until collision: 19.1 vs 22.4 | No clear effect |
| switch ancestral to nucleus sampling | Graph-backed model (0.6b) · BrickNet unconditional | steps until connectivity failure: 45.0 -> 94.1 | Improved |
| scale model size from 0.6b to 14b | Graph-backed model · BrickNet unconditional, nucleus sampling | steps until connectivity failure: 94.1 -> 96.9 | Improved |
| scale model size from 1.7b to 14b | BrickNet SFT · BrickNet text-conditioned | VQAScore: 0.59318309911395772 vs 0.60203790253581246 | Mixed |
| add unconditional pretraining before SFT (14b) (ablation) | BrickNet SFT · BrickNet held-out perplexity | perplexity: 1.298 -> 1.266 | Improved |
| add unconditional pretraining before SFT (0.6b) (ablation) | BrickNet SFT · BrickNet held-out perplexity | perplexity: 1.343 -> 1.298 | Improved |
| add more SFT data (None to Full, with full PT) (ablation) | BrickNet 14b · BrickNet held-out perplexity | perplexity: 1.300 -> 1.266 | Improved |
| add more PT data (None to Full SFT, quarter PT vs full PT) (ablation) | BrickNet 14b · BrickNet held-out perplexity | perplexity: 1.361 -> 1.300 | Improved |
Closest papers by meaning
- Generating Physically Stable and Buildable Brick Structures from Text (ICCV 2025 · similarity 0.92)
- BrickAnything: Geometry-Conditioned Buildable Brick Generation with Structure-Aware Tokenization (NeurIPS 2026 · similarity 0.91)
- Learning Stackable and Skippable LEGO Bricks for Efficient, Reconfigurable, and Variable-Resolution Diffusion Modeling (ICLR 2024 · similarity 0.85)
- Generative 3D Part Assembly via Part-Whole-Hierarchy Message Passing (CVPR 2024 · similarity 0.84)
- Imagine: Image-Guided 3D Part Assembly with Structure Knowledge Graph (AAAI 2025 · similarity 0.84)
- IKEA Manuals at Work: 4D Grounding of Assembly Instructions on Internet Videos (NeurIPS 2024 · similarity 0.84)
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.
