## Title: Dataset and model checkpoints - Inverse-design of nonlinear mechanical metamaterials via video denoising diffusion models

Version: v1.0.0
Release Date: 4th September 2023

---

### Dataset Description:

This dataset contains the data required to train the models related to 'Inverse-design of nonlinear mechanical metamaterials via video denoising diffusion models' and corresponding model checkpoints to sample new metamaterials for given stress-strain paths.
It is intended for research and academic purposes. The data was generated with Simulia Abaqus 2020.

---

### File Structure:

This data package includes the following:

1. 'pretrained.zip': The checkpoints of the model presented in the manuscript, used to generate the shown metamaterials.
2. 'lagrangian.zip': The main data file consisting of training and validation data for model training in the Lagrangian frame.
		     We provide the full field data as gifs, including the strain energy ('ener'), vertical stress components ('s_22'),
		     von-Mises stress ('s_mises'), binary topology ('topo'), and displacement fields ('u_1' and 'u_2').
		     The corresponding stress-strain curves are given in stress_strain_data.csv. To convert the uint8 gif pixel data into the
		     actual physical value, we provide the true physical value of a pixel value of '0' and '255' in frame_range_data.csv.
		     We provide further details on frame_range_data.csv in the next section.
3. 'eulerian.zip':   The data file consisting of training and validation data for model training in the Eulerian frame.
		     We provide the full field data as gifs, including the strain energy ('ener'), vertical stress components ('s_22'),
		     von-Mises stress ('s_mises'), and binary topology ('topo').
		     The corresponding stress-strain curves are given in stress_strain_data.csv. To convert the uint8 gif pixel data into the
		     actual physical value, we provide the true physical value of a pixel value of '0' and '255' in frame_range_data.csv.
		     Note that this dataset was used in preliminary studies but showed inferior performance to the Lagrangian frame.
		     We however include it in case other researchers might find it useful.
4. 'readme.txt':     This file, providing a brief overview of the dataset.

---

### Explanation of frame_range_data.csv:

All generated data is converted into the gif format for efficiency reasons. Since gif files store data in uint8 format, i.e., in integer values from 0 to 255,
we scale the full field data so that the minimum value (over all frames) is represented as a 0 and the maximum value as a 255. To convert this back to the physical
range, we provide the minimum and maximum values. In case the values are nonnegative (such as strain energy or von-Mises stress), we omit the minimum value, since it is 0.

In the Lagrangian frame, we store these values as follows:

| Col idx | Description         |
|---------|---------------------|
| 0       | Min 'u_1'           |
| 1       | Max 'u_1'           |
| 2       | Min 'u_2'           |
| 3       | Max 'u_2'           |
| 4       | Max 's_mises'       |
| 5       | Min 's_22'          |
| 6       | Max 's_22'          |
| 7       | Max 'ener'          |


In the Eulerian frame, we store these values as follows:

| Col idx | Description         |
|---------|---------------------|
| 0       | Max 's_mises'       |
| 1       | Min 's_22'          |
| 2       | Max 's_22'          |
| 3       | Max 'ener'          |

---

### Data Usage:

For further instructions on training the denoising algorithm and generating new metamaterials, please follow the instructions given in the repository (https://github.com/jhbastek/VideoMetamaterials).
This dataset is provided "as is" and for academic and research purposes. Please consider citing our manuscript in case this dataset is of use in your research.

---

### Contact Information:

For any questions related to this dataset, please read the manuscript and corresponding supplementary information, or reach out to the data curator:

Name: Jan-Hendrik Bastek
Email: jbastek@ethz.ch

---

### Disclaimer:

This dataset is provided without any express or implied warranties. The user assumes full responsibility for any outcomes that arise from using this data.

---