Structure-Preserving 3D Garment Modeling with Neural Sewing Machines
Authors: Xipeng Chen, Guangrun Wang, Dizhong Zhu, Xiaodan Liang, Philip Torr, Liang Lin
NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments demonstrate that the proposed NSM is capable of representing 3D garments under diverse garment shapes and topologies, realistically reconstructing 3D garments from 2D images with the preserved structure, and accurately manipulating the 3D garment categories, shapes, and topologies, outperforming the state-of-the-art methods by a clear margin. |
| Researcher Affiliation | Academia | 1 Sun Yat-sen University 2 University of Oxford |
| Pseudocode | No | The paper does not contain any structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | The code of NSM is in the supplemental. |
| Open Datasets | Yes | Here we use the largest 3D garment dataset with sewing patterns introduced in [15]. |
| Dataset Splits | No | The paper states: 'We use the first 90% part of the data for each base type as the training set and the remaining as the test set.' It does not explicitly mention a separate validation split or how it was used for hyperparameter tuning or early stopping. |
| Hardware Specification | Yes | Our experiments are conducted on Ubuntu System, Intel(R) Core(TM) i5-8400 CPU(2.80GHz), GTX 2080. |
| Software Dependencies | No | The paper mentions implementing models with CNNs and a U-Net architecture but does not specify versions for any key software components or libraries (e.g., PyTorch, TensorFlow, or specific Python libraries). |
| Experiment Setup | Yes | The weights {αrec, αinn, αint, αnom} in the training loss L are set to {1, 10 3, 10 4, 10 2}. The number of PCA components h is set to 12... The basic panel group number are set to 10 and the total panel number is set to 33. The scaling s between the 3D coordinate and the UV coordinate is set to 1.5. The width W and height H of the UV-position maps and the mask maps are set as 128. Our model is trained with a GTX 2080 GPU with the learning rate as 1e 3 and the batch size as 8 for 40 epochs. |