Non-Iterative Inverse Design for 4D Printing via Fabrication-Ready Discrete Material Programming
Description
This dataset contains comprehensive resources for 4D printing inverse design research, including segmentation training data and pre-trained model weights. The package encompasses two main components: 1. Segmentation Dataset (morphpad_16_labels_2voids) The dataset includes semantic segmentation image data with pixel-level annotations for 4D printing design optimization. It contains RGB images paired with 17-class segmentation masks (16 material/orientation classes + 1 void class). Data is organized across three different synthesis methods: Gaussian noise synthesis Patch-based synthesis Perlin noise synthesis Each synthesis method contains paired images/ and labels/ subdirectories. The id2label.json file provides the mapping between class IDs and semantic labels. 2. Model Checkpoint (model_vpp_tao2026) Pre-trained semantic segmentation model weights based on SegFormer architecture (HuggingFace transformers). The checkpoint contains: config.json (model configuration) model.safetensors (pre-trained weights) optimizer.pt, scheduler.pt (training states) trainer_state.json (training metadata and evaluation metrics) Use Cases: Training and evaluating image segmentation models Material and orientation classification 4D printing design optimization research Direct inference on fabrication-relevant patterns Technical Details: Framework: PyTorch / HuggingFace Transformers Input: Coordinate-rendered RGB images Output: 17-class segmentation masks Training data: morphpad_16_labels_2voids dataset Dependencies: transformers>=4.20.0, torch>=1.9.0, pillow>=8.0.0 Authors: Quanqing Tao, Jiahai Ma, Zongxin Hu, Haitao Ye, Deyong Sun, Winston Wai Shing Ma, Haoming Mo, Qiguang He, Xu Song Affiliations: Department of Mechanical and Automation Engineering, Chinese University of Hong Kong, Shatin, Hong Kong SAR Research Center of Intelligent Manufacturing Technology, Beijing Institute of Technology, Zhuhai, Guangdong 519088, China Journal: Virtual and Physical Prototyping Year: 2026 License: CC BY 4.0 Corresponding Authors: qiguanghe@cuhk.edu.hk, xsong@cuhk.edu.hk