ReWeaver-Code

KO | EN

gitlite — search

#javascript #python #hacktoberfest #react #ai #typescript #llm #go #golang #android #machine-learning #rust #deep-learning #linux

★ 27 Open GitHub ↗

[CVPR 2026] It's the official repository of "ReWeaver: Towards Simulation-Ready and Topology-Accurate Garmemt Reconstruction".

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# ReWeaver: Towards Simulation-Ready and Topology-Accurate Garment Reconstruction

## 🌟 Overview

![ReWeaver Teaser](https://raw.githubusercontent.com/SII-LiMing/ReWeaver-Code/main/asset/Teaser.png)

**ReWeaver** reconstructs structured 3D garments and 2D sewing patterns from sparse multi-view RGB images. The method jointly predicts 3D curves, 3D patches, patch-curve connectivity, and flattened 2D panel edges, producing garment assets that are better aligned with downstream simulation and editing pipelines.

---

## Get Started with ReWeaver

### 🛠️ Preparation

1. **💻 Environment.**  
   The current codebase was developed with Python `3.12.9` and CUDA `12.4`.

```bash
git clone https://github.com/SII-LiMing/ReWeaver-Code.git
cd ReWeaver-Code

conda create -n reweaver python=3.12.9
conda activate reweaver

pip install ipdb tqdm tyro omegaconf trimesh wandb scipy tensorboard matplotlib pillow scikit-image

# for cuda 12.4, you can install torch according your cuda version.
pip install torch==2.6.0 torchvision==0.21.0 torchaudio==2.6.0 --index-url https://download.pytorch.org/whl/cu124

pip install "git+https://github.com/facebookresearch/pytorch3d.git@stable"
```

2. **📥 Pretrained weights.**  
  Download the pre-trained weights from Hugging Face: [ReWeaver](https://huggingface.co/SII-LiMing/ReWeaver/tree/main). We provide two versions:

- `tileable/`: trained on data rendered with tileable textures
- `GCD_ori/`: trained on data rendered with the original GCD textures

Each version contains:

```text
complex_stitch.pth
flatten.pth
img_encoder.pth
```

3. **⚙️ Config files.**  
   Main configs are stored in [`configs/`](configs/). Before training or testing, you should update:

- `train_data.root`, `eval_data.root`, `test_data.root`
- `save_dir`, `exp_name`
- `complex_model_path`, `flatten_model_path`, `img_encoder_model_path` for evaluation

---

### 📂 Training Dataset

We train ReWeaver on **GCD-TS**, a textured multi-view garment dataset introduced in the paper. The public dataset link is:

- 🤗 https://huggingface.co/datasets/SII-LiMing/ReWeaver-GCD-TS

For the current codebase, each GCD-style sample directory is expected to contain:

- RGB images under:
  - `render_output/rgb/` when `texture_type: tileable`
  - `render_output_ori_texture/rgb/` when `texture_type: ori`
- a panel annotation file:
  - `<sample_name>_2d_panel.json`

The loader scans each direct child folder under `train_data.root`, `eval_data.root`, or `test_data.root` unless `samples` is set to a text file listing selected samples.

---

### 🚀 Training

The current training / evaluation entrypoint is [`main.py`](main.py).  
Pass the YAML config from the command line, for example:

- [`configs/train_tileable.yaml`](configs/train_tileable.yaml)
- [`configs/train_gcd.yaml`](configs/train_gcd.yaml)

Then launch distributed training:

```bash
CUDA_VISIBLE_DEVICES=0 torchrun --standalone --nproc_per_node=1 main.py --config configs/train_tileable.yaml
```

---

### 🧪 Evaluation
Recommended test configs in this repo:

- [`configs/eval_gcd.yaml`](configs/eval_gcd.yaml)
- [`configs/eval_tileable.yaml`](configs/eval_tileable.yaml)

For example:

```bash
CUDA_VISIBLE_DEVICES=0 torchrun --standalone --nproc_per_node=1 main.py --config configs/eval_tileable.yaml
```

Predictions are saved as compressed `.npz` files under:

```text
<save_dir>/<exp_name>/pred/test/<sample_name>/<sample_name>.npz
```

---

## 💖 Acknowledgement

This repository builds on open-source tools and prior work from the community. Many thanks to the authors for sharing their code and resources.

- 🔗 [vggtencoder/](https://arxiv.org/abs/2503.11651)
- 🔗 [ComplexGen](https://arxiv.org/abs/2205.14573)

---

## 📜 Citation

If you find this repository useful in your research, please cite:

```bibtex
@inproceedings{li2026reweaver,
  title={ReWeaver: Towards Simulation-Ready and Topology-Accurate Garment Reconstruction},
  author={Li, Ming and Shan, Hui and Zheng, Kai and Shen, Chentao and Liu, Siyu and Fu, Yanwei and Chen, Zhen and Huang, Xiangru},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  year={2026},
  url={https://sii-liming.github.io/ReWeaver/}
}
```

ReWeaver-Code

Explore Similar Repositories

Was this content helpful?

Recent Feedback

Download README