Towards automatic digital garment assembly

Alejandro Rodriguez

cs.CVOct 12, 2026

Towards automatic digital garment assembly

Authors: Alejandro Rodriguez

Organizations: SEDDI.

Abstract

Creating digital garments from existing two-dimensional patterns remains a time-consuming process. Although patterns are widely available in digital form, their assembly in garment CAD software involves manual definition of sewing operations and arrangement of the pattern panels for drape simulation. We present two learning-based approaches that automate both tasks using the 2D vector pattern as their sole input. Our automatic sewing model operates at the segment level, leveraging a Transformer encoder with multiple prediction heads. Our novel architecture supports many-to-many, ordered segment correspondences, as well as hems, overlapping seams needed for multilayer garments, and construction properties such as seam allowances and stitch types, required for the finished virtual garment but neglected by previous methods. A complementary panel-level model predicts body-region assignments, layers, and placement coordinates to initialize the drape simulation. We evaluate both models on a curated dataset of 4,035 professionally assembled garment instances. The sewing model exactly recovers all seam and hem topologies of 63.3% of test garments, while the arrangement model meets the specified placement criteria for all panels in 66.2% of test garments. We also integrate both methods into garment CAD software, demonstrating automatic assembly of successfully predicted garments.

Figures & tables

Fig. 1: Application of our proposed solutions integrated into an apparel CAD system. The input 2D pattern, shown at the top left, is processed by our automatic sewing and arrangement models to assemble the garment into a simulation-ready configuration. After draping, the True Seams process [ 1 ] creates the final, detailed geometry. The entire process is automatic, going from 2D pattern to finished 3D asset without manual adjustments required.

Fig. 2: Example of a finished garment with hems and facing seams. The collar has two layers of fabric joined together by facing seams before being sewn to the torso panels with a regular seam. The sleeves and the shirt bottom are finished with hems.

Fig. 3: Proposed automatic sewing model architecture. Per-segment features

X

, including panel-level features, are input to the model. They are projected into embeddings

E

and refined by a Transformer encoder with full self-attention. Specialized heads then process the enriched embeddings

H

to reconstruct the topology and properties of the different sewing operations.

Fig. 4: Toy pattern example with a single multi-segment seam, showing the corresponding target parity and order matrices encoding the seam topology.

Fig. 5: Proposed architecture for the automatic arrangement model. The sampled boundaries of the target garment panels are processed by a 1D cyclic ResNet to produce per-panel embeddings

E_{p}

. Positional embeddings

P

are computed and added to enrich the panel embeddings

H_{p}

. They get further refined with contextual awareness by a Transformer encoder with full self-attention. The final embeddings

H_{c}

are then passed through simple linear projections to generate the per-panel arrangement configurations.

Fig. 6: Arrangement tool provided by the CAD software. On the left side, a 2D space allows users to assign the different pattern panels to body regions. These assignments map the panel geometries to the 3D space on the right, wrapping them around the body, properly arranging them for a successful drape simulation.

Fig. 7: Examples of garments of different complexity in our dataset. The pink tank, shown at the top right, has a very simple pattern with 2 panels, 4 regular seams and 4 hems. The navy jacket, shown at the bottom right, has a more complex pattern with 47 panels, 45 regular seams and 23 facing seams.

	Average	Min	Max	Total
Garments	-	-	-	4,035
Panels	8.68	2	52	35,038
Regular seams	13.19	4	49	53,220
Facing seams	1.56	0	25	6,312
Hems	2.54	0	14	10,243

TABLE I: Dataset statistics on garment panels and sewing operations.

	Perfect recovery (%)
Method	Garments	R. seams	F. seams	Hems
Ours	63.28	91.52	77.68	86.12
Neighbor att.	48.31	87.06	71.10	83.18
Extra Bézier feat.	58.45	90.72	75.61	86.73
Negative mining	60.63	91.46	78.29	85.36

TABLE II: Results of the automatic sewing experiments on the test set, with perfect recovery percentages for whole garments and for regular seams, facing seams and hems. Corresponding results for the ablation experiments are also reported.

Task	Acc.	Prec.	Recall	F1	MAE (cm)
R. $S_{parity}$	99.79	95.74	95.97	95.85	-
R. $S_{order}$	96.63	96.63	96.63	96.63	-
F. $S_{parity}$	99.93	93.12	90.93	92.01	-
F. $S_{order}$	99.15	99.15	99.15	99.15	-
Hem segment	99.13	96.10	90.14	93.03	-
Fold side	85.94	85.93	85.94	85.93	-

TABLE III: Metrics corresponding to the evaluation outputs of the different automatic sewing model heads. We report accuracy (acc.), precision (prec.), recall and F1 for regular (R.), facing (F.) and hem classification tasks. We also report mean absolute error (MAE) for seam allowance regression. Properties are evaluated per labeled segment.

Fig. 8: Examples of predicted sewing operations and panel arrangements in the CAD system. Each example shows the sewn 2D pattern (left) and the arrangement around the 3D body (right). Top row: correct sewing operations and arrangements. Middle row: missing seams. Bottom-row: incorrect arrangement assignments. Mistakes are circled in red.

Method	Garments (%)	Panels (%)
Ours	66.18	82.61
w/o Transformer	34.29	67.68

TABLE IV: Results of the automatic arrangement experiments on the test set, listing percentage of correctly arranged panels and garments. Results for the ablation experiment are also reported.

Task	Acc.	Prec.	Recall	F1	E. distance (cm)
Body p.	94.94	94.77	94.94	94.72	-
Layer	89.28	89.61	89.28	89.40	-
Pos.	-	-	-	-	3.76

TABLE V: Metrics corresponding to the evaluation outputs of the different automatic arrangement model heads. We report accuracy (acc.), precision (prec.), recall and F1 for body region and layer tasks, and Euclidean distance error for arrangement position.

AutoSew: A Geometric Approach to Stitching Prediction with Graph Neural Networks
Feb 25, 2026Pablo Ríos-Navarro, Elena Garces, Jorge Lopez-MorenoManufacturing Graph Matching
Universidad Rey Juan Carlos, Madrid, Spain · Adobe Research, France
Stitched Embeddings: A Unified Latent Space for 3D Garments and 2D Patterns
Jul 1, 2026Andrea Sanchietti, Riccardo Marin, Bharat Lal Bhatnagar +23D Reconstruction 3D Garment Reconstruction
University of Tübingen, Germany · Tübingen AI Center · Technical University of Munich, Germany +3
InverseDraping: Recovering Sewing Patterns from 3D Garment Surfaces via BoxMesh Bridging
Apr 3, 2026Leyang Jin, Zirong Jin, Zisheng Ye +43D Garment Reconstruction 3D Reconstruction
Mohamed bin Zayed University of Artificial Intelligence · The Chinese University of Hong Kong, Shenzhen · School of Science and Engineering, The Chinese University of Hong Kong, Shenzhen, FNii-Shenzhen, and Guangdong Provincial Key Laboratory of Future Networks of Intelligence +1

Feb 25, 2026J/K move · Enter open · S save

AutoSew: A Geometric Approach to Stitching Prediction with Graph Neural Networks

Pablo Ríos-Navarro, Elena Garces, Jorge Lopez-Moreno

Universidad Rey Juan Carlos, Madrid, Spain · Adobe Research, France

Manufacturing Graph Matching

Open paper AskPDF arXiv

Automating garment assembly from sewing patterns remains a significant challenge due to the lack of standardized annotation protocols and the frequent absence of semantic cues. Existing methods often rely on panel labels or handcrafted heuristics, which limit their applicability to real-world, non-conforming patterns. We present AutoSew, a fully automatic, geometry-based approach for predicting stitch correspondences directly from 2D pattern contours. AutoSew formulates the problem as a graph matching task, leveraging a Graph Neural Network to capture local and global geometric context, and employing a differentiable optimal transport solver to infer stitching relationships-including multi-edge connections. To support this task, we update the GarmentCodeData dataset modifying over 18k patterns with realistic multi-edge annotations, reflecting industrial assembly scenarios. AutoSew achieves 96% F1-score and successfully assembles 73.3% of test garments without error, outperforming existing methods while relying solely on geometric input. Our results demonstrate that geometry alone can robustly guide stitching prediction, enabling scalable garment assembly without manual input.

Abstract

Explore similar work

AutoSew: A Geometric Approach to Stitching Prediction with Graph Neural Networks

Stitched Embeddings: A Unified Latent Space for 3D Garments and 2D Patterns

InverseDraping: Recovering Sewing Patterns from 3D Garment Surfaces via BoxMesh Bridging