Computer-Aided Design (CAD) generative modeling has a strong and long-term application in the industry. Recently, the parametric CAD sequence as the design logic of an object has been widely mined by sequence models. However, the industrial CAD models, especially in component objects, are fine-grained and complex, requiring a longer parametric CAD sequence to define. To address the problem, we introduce Mamba-CAD, a self-supervised generative modeling for complex CAD models in the industry, which can model on a longer parametric CAD sequence. Specifically, we first design an encoder-decoder framework based on a Mamba architecture and pair it with a CAD reconstruction task for pre-training to model the latent representation of CAD models; and then we utilize the learned representation to guide a generative adversarial network to produce the fake representation of CAD models, which would be finally recovered into parametric CAD sequences via the decoder of MambaCAD. To train Mamba-CAD, we further create a new dataset consisting of 77,078 CAD models with longer parametric CAD sequences. Comprehensive experiments are conducted to demonstrate the effectiveness of our model under various evaluation metrics, especially in the generation length of valid parametric CAD sequences. The code and dataset can be achieved from https://github.com/Sunny-Hack/Code-for-Mamba-CAD-AAAI-2025-.
Figures & tables
Figure 1: The gallery of CAD models created by Mamba-CAD, including visual illustrations of the created CAD models from CAD sequence reconstruction, completion, and random generation.
Figure 2: Common CAD commands discussed in this paper. < SOL > and < EOS > are two symbols for representing the start of one loop and the end of a whole parametric CAD sequence, respectively. f is a counter-clockwise flag.
Figure 3: The overview of our Mamba-CAD architecture. The whole pipeline can be divided into three parts which marked in three different colored line-arrows. (Pre-training stage ⟶ ) : We leverage a task of CAD sequence reconstruction to pre-train the encoder and decoder of Mamba-CAD. Given a mini-batch of CAD-models, their corresponding parametric CAD sequences would be first discretized into friendly network representations (Sec. CAD sequence representation CAD sequence representation ). Next, these representations keep going forward Fusion Embedding, Encoder, and Decoder of Mamba-CAD (Sec. Mamba-CAD Mamba-CAD ) to recover predicted parametric CAD sequences. ( Training stage ⟶ ) : When the pre-training stage is complete, the encoder parameters of Mamba-CAD would be frozen to produce the latent representation K that is further used to guide a 1-D Latent-GAN training. ( Generation stage ⟶ ) : Once the 1-D Latent GAN is well-trained, it enables to randomly sample the noise from a standard gaussian distribution and further feed it into Generator to produce the fake latent representation F , which would further undergo the decoder of Mamba-CAD to recover the parametric CAD sequence.
Dataset
TN
AL
[1-10]
[11-25]
[26-40]
[41-60]
[60-128]
DeepCAD
178,238
15
44.58%
38.73%
11.38%
5.31%
-
Mamba-CAD
77,078
38
-
41.25%
26.60%
16.13%
16.02%
Table 1: The statistic comparison of between DeepCAD dataset and Our Mamba-CAD dataset, including total number (TN), average length (AL) of parametric CAD sequences, and the length distribution.
Figure 4: Visual illustrations of CAD sequence reconstruction. GT represents Ground Truth. ∅ denotes the CAD sequence is invalid, leading to the failure of constructing a 3D shape. Please enlarge 200% for a clear view.
Method
Ac↑
Ap↑
MCD ↓
IR ↓
AL ↑
L ≥ 60 ↑
SR ↑
DeepCAD
96.49
87.38
2.97
14.83%
32.52
51.82%
86.58%
DeepCAD (++)
96.92
87.68
2.81
14.05%
32.49
50.85%
87.49%
TM-CAD
95.73
85.48
4.82
15.85%
31.21
50.31%
85.61%
Mamba-CAD
99.96
99.93
0.64
5.41%
38.91
85.18%
95.43%
Table 2: The comparison of Mamba-CAD and variants in CAD sequence reconstruction. MCD is multiplied by 103 . Ac , Ap , and IR are multiplied by 100%. L ≥ 60: the ratio of valid reconstructed CAD sequences (Length ≥ 60) in total CAD sequences (Length ≥ 60) within test set.
Method
Ac↑
Ap↑
MCD ↓
IR ↓
AL ↑
L ≥ 60 ↑
SR ↑
DeepCAD
83.75
69.77
79.08
42.36%
28.72
19.11%
59.19%
DeepCAD (++)
85.95
71.27
69.68
40.19%
29.78
20.74%
61.12%
TM-CAD
80.75
67.71
82.34
44.34%
28.26
17.79%
57.23%
HNC-CAD
-
-
-
-
30.51
42.89%
74.36%
Mamba-CAD
99.85
98.91
1.03
8.03%
35.43
79.46%
92.84%
Table 3: Quantitative results of CAD sequence completion.
Method
Ac↑
Ap↑
MCD ↓
IR ↓
SR ↑
DeepCAD
95.23
85.58
10.97
10.78%
90.34%
DeepCAD (++)
96.79
87.72
8.93
9.35%
91.22%
TM-CAD
90.94
83.85
12.46
12.37%
87.61%
Mamba-CAD
99.21
98.21
1.31
3.72%
96.51%
Table 4: The quantitative results of generalization ability in Fusion360 reconstruction dataset.
Figure 5: Visualizations of random generation of parametric CAD sequences.
Method
MMD ↓
JSD ↓
COV ↑
Unique ↑
Novel ↑
AL ↑
SR ↑
DeepCAD
3.51
8.93
63.77%
83.41
51.94
27.78
62.76%
SkexGen
2.96
7.39
65.89%
93.33
60.47
29.18
63.49%
HNC-CAD
2.65
6.07
70.43%
92.18
55.32
31.36
68.84%
Mamba-CAD
1.94
4.28
75.61%
94.73
63.25
36.75
74.13%
Table 5: The quantitative results of random generation of parametric CAD sequences. MMD, JSD, Unique, and Novel are all multiplied by 102 .
Parametric CAD modeling represents 3D designs as editable, constraint-driven construction histories and develops methods to generate, reconstruct, and revise them.
Parametric Computer-Aided Design (CAD) is fundamental to modern 3D modeling, yet existing methods struggle to generate long command sequences, especially under complex geometric and topological dependencies. Transformer-based architectures dominate CAD sequence generation due to their strong dependency modeling, but their quadratic attention cost and limited context windowing hinder scalability to long programs. We propose GeoFusion-CAD, an end-to-end diffusion framework for scalable and structure-aware generation. Our proposal encodes CAD programs as hierarchical trees, jointly capturing geometry and topology within a state-space diffusion process. Specifically, a lightweight C-Mamba block models long-range structural dependencies through selective state transitions, enabling coherent generation across extended command sequences. To support long-sequence evaluation, we introduce DeepCAD-240, an extended benchmark that increases the sequence length ranging from 40 to 240 while preserving sketch-extrusion semantics from the ABC dataset. Extensive experiments demonstrate that GeoFusion-CAD achieves superior performance on both short and long command ranges, maintaining high geometric fidelity and topological consistency where Transformer-based models degrade. Our approach sets new state-of-the-art scores for long-sequence parametric CAD generation, establishing a scalable foundation for next-generation CAD modeling systems. Code and datasets are available at GitHub.
Xiaolei Zhou, Chuangjie Fang, Jie Wu +3
Zhejiang University of Technology · Hangzhou International Innovation Institute, Beihang University
Computer-Aided Design (CAD) relies on structured and editable geometric representations, yet existing generative methods are constrained by small annotated datasets with explicit design histories or boundary representation (BRep) labels. Meanwhile, millions of unannotated 3D meshes remain untapped, limiting progress in scalable CAD generation. To address this, we propose DreamCAD, a multi-modal generative framework that directly produces editable BReps from point-level supervision, without CAD-specific annotations. DreamCAD represents each BRep as a set of parametric patches (e.g., Bézier surfaces) and uses a differentiable tessellation method to generate meshes. This enables large-scale training on 3D datasets while reconstructing connected and editable surfaces. Furthermore, we introduce CADCap-1M, the largest CAD captioning dataset to date, with 1M+ descriptions generated using GPT-5 for advancing text-to-CAD research. DreamCAD achieves state-of-the-art performance on ABC and Objaverse benchmarks across text, image, and point modalities, improving geometric fidelity and surpassing 75% user preference. Code and dataset will be publicly available.
Mohammad Sadil Khan, Muhammad Usama, Rolandos Alexandros Potamias +4
DFKI, Germany · RPTU Kaiserslautern, Germany · Imperial College London, United Kingdom +1
Multi-modal approaches used for 3D CAD generation require substantial computational resources, necessitating efficient training. To address this, we propose GuideCAD, which leverages semantically rich visual-textual representations having only a small number of trainable parameters to generate 3D CAD models. Specifically, GuideCAD uses a mapping network that converts image embeddings into prefix embeddings, enabling a pretrained large language model to integrate visual and textual information. As a result, a transformer-based decoder predicts the construction sequence using the visual-textual embeddings in order to generate the 3D CAD model. For experimental evaluation, we construct a new dataset, referred to as GuideCAD, which consists of text-image pairs. Each pair includes a text prompt that represents a 3D CAD construction sequence and its corresponding 3D CAD image. Our experimental results show that GuideCAD generates comparably high-quality 3D CAD models while using approximately four times fewer parameters and achieving twice the training efficiency compared to fine-tuning approaches. We have released the source code and dataset for our method at: https://github.com/mskimS2/GuideCAD
Minseong Kim, Jinyeong Park, Sungho Park +1
Convergence Research Center for Insect Vectors Incheon, South Korea · Department of Computer Science and Engineering Incheon National University Incheon, 22012, South Korea · Department of Computer Science and Engineering Center for Brain-Machine Interface Incheon National University Incheon, South Korea