Research map / Vision-language and multimodal / Multimodal LLMs
BrepCoder: A Unified Multimodal Large Language Model for Multi-task B-rep Reasoning
Mingi Kim, Yongjun Kim, Jungwoo Kang, Hyungki Kim
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 Phase 1 B-rep-code alignment before LLM integration (vs end-to-end training) (ablation) | BrepCoder · DeepCAD reverse engineering | Median Chamfer Distance (x1e-3): 124.18 -> 0.464 | Improved |
| train encoder from scratch instead of segmentation pre-trained weights (ablation) | BrepCoder · DeepCAD reverse engineering | Median Chamfer Distance (x1e-3): 0.499 -> 0.464 | Improved |
| use UV-Net instead of AAGNet encoder (finer 10x10 UV sampling) (ablation) | BrepCoder · DeepCAD reverse engineering | Median Chamfer Distance (x1e-3): 3.99 -> 0.464 | Improved |
| use CoCa (contrastive plus captioning loss) instead of CLIP alignment (ablation) | BrepCoder · DeepCAD reverse engineering | ACC_param: 78.09 -> 82.01 | Improved |
| freeze the B-rep encoder when integrating with LLM (ablation) | BrepCoder · DeepCAD reverse engineering | Median Chamfer Distance (x1e-3): 0.516 -> 0.464 | Improved |
| scale LLM backbone from 0.5B to 1.5B (ablation) | BrepCoder · DeepCAD reverse engineering | Median Chamfer Distance (x1e-3): 0.612 -> 0.464 | Improved |
| swap Qwen-2.5 1.5B backbone for Llama-3.2 1B (ablation) | BrepCoder · DeepCAD reverse engineering | Median Chamfer Distance (x1e-3): 0.464 -> 0.467 | No clear effect |
| use 3-layer projector instead of 2-layer (ablation) | BrepCoder · DeepCAD reverse engineering | ACC_param: 80.15 -> 82.01 | Improved |
| use 3-layer projector instead of 1-layer (ablation) | BrepCoder · DeepCAD reverse engineering | ACC_param: 80.66 -> 82.01 | Improved |
| add Stage 1 reverse engineering pre-training (ablation) | BrepCoder · CAD-QA (SGP-Bench) | accuracy (%): 71% -> 79% | Improved |
| add Stage 1 reverse engineering pre-training (ablation) | BrepCoder · Error correction (DeepCAD) | Median Chamfer Distance (x1e-3): 0.405 -> 0.335 | Improved |
| add Stage 1 reverse engineering pre-training (ablation) | BrepCoder · Error correction (DeepCAD) | ACC_cmd: 99.43 -> 99.08 | Hurt |
Closest papers by meaning
- cadrille: Multi-modal CAD Reconstruction with Reinforcement Learning (ICLR 2026 · similarity 0.87)
- CAD-Coder: Text-to-CAD Generation with Chain-of-Thought and Geometric Reward (NeurIPS 2025 · similarity 0.87)
- From Intent to Execution: Multimodal Chain-of-Thought Reinforcement Learning for Precise CAD Code Generation (AAAI 2026 · similarity 0.86)
- ReCAD: Reinforcement Learning Enhanced Parametric CAD Model Generation with Vision-Language Models (AAAI 2026 · similarity 0.85)
- CAD-GPT: Synthesising CAD Construction Sequence with Spatial Reasoning-Enhanced Multimodal LLMs (AAAI 2025 · similarity 0.85)
- Text-to-CAD Generation Through Infusing Visual Feedback in Large Language Models (ICML 2025 · similarity 0.85)
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.
