The ABC Dataset is a collection of one million Computer-Aided Design (CAD) models for research of geometric deep learning methods and applications. Each model is a collection of explicitly parametrized curves and surfaces, providing ground truth for differential quantities, patch segmentation, geometric feature detection, and shape reconstruction. Sampling the parametric descriptions of surfaces and curves allows generating data in different formats and resolutions, enabling fair comparisons for a wide range of geometric learning algorithms.

4D-DRESS is the first real-world 4D dataset of human clothing, capturing 64 human outfits in more than 520 motion sequences. These sequences include a) high-quality 4D textured scans; for each scan, we annotate b) vertex-level semantic labels, thereby obtaining c) the corresponding garment meshes and fitted SMPL(-X) body meshes. Totally, 4D-DRESS captures dynamic motions of 4 dresses, 28 lower, 30 upper, and 32 outer garments. For each garment, we also provide its canonical template mesh to benefit the future human clothing study.

PasticineLab is a differentiable physics benchmark, which includes a diverse collection of soft body manipulation tasks. In each task, the agent uses manipulators to deform the plasticine into the desired configuration. The underlying physics engine supports differentiable elastic and plastic deformation using the DiffTaichi system, posing many under-explored challenges to robotic agents.

CMD is a publicly available collection of hundreds of thousands 2D maps and 3D grids containing different properties of the gas, dark matter, and stars from more than 2,000 different universes. The data has been generated from thousands of state-of-the-art (magneto-)hydrodynamic and gravity-only N-body simulations from the CAMELS project. Each 2D map and 3D grid has a set of labels associated to it: 2 cosmological parameters characterizing fundamental properties of the Universe, and 4 astrophysical parameters parametrizing the strength of astrophysical processes such as feedback from supernova and supermassive black-holes. The main task this dataset was designed is to perform a robust inference on the value of the cosmological parameters from each map and grid. The data itself was generated from two completely different set of simulations, and it is not obvious that training one model on one will work when predicting on the other. Since simulations of the real Universe may never be perfect, this dataset provides the data to tackle this problem. Solving this problem will help cosmologists to constrain the value of the cosmological parameters with the highest accuracy and therefore unveil the mysteries of our Universe. CMD can also be used for many other tasks, such as field mapping and super-resolution.

Numerical simulations of Earth's weather and climate require substantial amounts of computation. This has led to a growing interest in replacing subroutines that explicitly compute physical processes with approximate machine learning (ML) methods that are fast at inference time. Within weather and climate models, atmospheric radiative transfer (RT) calculations are especially expensive. This has made them a popular target for neural network-based emulators. However, prior work is hard to compare due to the lack of a comprehensive dataset and standardized best practices for ML benchmarking. To fill this gap, we build a large dataset, ClimART, with more than \emph{10 million samples from present, pre-industrial, and future climate conditions}, based on the Canadian Earth System Model. ClimART poses several methodological challenges for the ML community, such as multiple out-of-distribution test sets, underlying domain physics, and a trade-off between accuracy and inference speed.

DrivAerNet is a large-scale, high-fidelity CFD dataset of 3D industry-standard car shapes designed for data-driven aerodynamic design. It comprises 4000 high-quality 3D car meshes and their corresponding aerodynamic performance coefficients, alongside full 3D flow field information. It includes: - CFD Simulation Data: The raw dataset, including full 3D pressure, velocity fields, and wall-shear stresses, computed using 8-16 million mesh elements has a total size of $\sim$ 16TB. - Curated CFD Simulations: For ease of access and use, a streamlined version of the CFD simulation data is provided, refined to include key insights and data, reducing the size to $\sim$ 1TB. - 3D Car Meshes: A total of 4000 designs, showcasing a variety of conventional car shapes and emphasizing the impact of minor geometric modifications on aerodynamic efficiency. The 3D meshes and aerodynamic coefficients $\sim$ 84GB. - 2D slices include the car's wake in the $x$-direction and the symmetry plane in the $y$-direction $\sim$ 12GB.

The dataset is a .h5 file comprised of entries with keys of the form (n,m), denoting the dimensions of the system matrix on which the simulations have been performed. The value of each key are two arrays, one to store the number of iterations needed to terminate the process for each simulation, and the other for the number of elements present at the terminal state of each simulation. Thus, given the maximum number of elements n*m in each system, the estimated percolation threshold can be computed by averaging the ratios between the elements at each terminal state and the system size. Overall, 207950010 simulations have been performed. And, this dataset was used to perform a complexity analysis on: https://arxiv.org/abs/2410.11874

Dataset of low fidelity resolutions of the RANS equations over airfoils.

The dataset consists of high-resolution three-dimensional (3D) turbulent flow simulations. It captures intricate vortex structures caused by a variety of shapes within a channel flow environment. The dataset is generated using OpenFOAM in large eddy simulation (LES) mode, ensuring the preservation of detailed turbulent characteristics across all spatial scales. The dataset contains 45 simulations consisting of 5000 flow snapshots each at a temporal resolution of 0.1ms, for a total physical runtime of 0.5s per simulation. The channel geometry is 0.4x0.1x0.1m discretizted into a 192x48x84 regular grid. The inflow velocity is 20 meters per second. Each simulation contains a unique turbulent flow caused by the specific shape in the channel. This dataset is ideally suited for machine learning researchers aiming to experiment with surrogate models, generative models, and applications in CFD. It can be used to train neural networks to predict turbulent flow predictions, simulate realistic flow fields efficiently, and evaluate models that aim to replicate the properties of turbulent kinetic energy distribution accurately. Moreover, it provides a rigorous testbed for developing generative models that can bypass computationally expensive CFD simulations, thereby prompting innovations in faster, model-driven fluid dynamics simulations for industrial and academic utilizations.

This dataset contains data of 125 1-hour simulations of ship motion during various sea states performing random maneuvers in 4 degrees of freedom (surge-sway-yaw-roll). The original ship is a patrol ship developed by Perez et al. [1]. We have extended it with a set of two symmetrically placed rudder propellers. Additionally, we simulate wind forces according to Isherwood's wind model [2]. Wind-induced waves are generated with the JONSWAP spectrum [3] and the corresponding wave forces are then computed using wave force response amplitude operators (ROA). Implementations of the ship model, Isherwood's wave model, wave force ROAs and the JONSWAP spectrum can be found in the Marine Systems Simulator toolbox by Fossen and Perez [4]. The dataset is split into a routine operation set (96 hours) and into an Out-Of-Distribution (OOD) set (29 hours). The routine operation set is split into train-validation-test with a 60-10-30 split, while the OOD set is used solely for testing. The dataset is used for the evaluation of nonlinear system identification methods for multi-step predictions. The following inputs and outputs are considered for the identification problem. Inputs are the shaft speeds of both propellers, their azimut angles, wind angle of attack, and wind speed. Measured states or outputs are surge velocity, sway velocity and roll rate, as well as yaw angle and roll angle. Please see the README.txt file for details regarding the file structure of this dataset and a description of the variables in the .tab files. This research is funded by Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) under Germany's Excellence Strategy - EXC 2075 - 390740016. We acknowledge the support by the Stuttgart Center for Simulation Science (SimTech). [1] T. Perez, A. Ross, and T. I. Fossen, “A 4-DOF SIMULINK model of a coastal patrol vessel for manoeuvring in waves,” in IFAC MCMC, 2006. [2] R. M. Isherwood, “Wind resistance of merchant ships,” The Royal Institution of Naval Architects, 1972. [3] K. Hasselmann, T. Barnett, E. Bouws, H. Carlson, D. Cartwright, K. Enke, J. Ewing, H. Gienapp, D. Hasselmann, P. Kruseman, A. Meerburg, P. Muller, D. Olbers, K. Richter, W. Sell, and H. Walden, “Measurements of wind-wave growth and swell decay during the joint north sea wave project (JONSWAP),” Deut. Hydrogr. Z., vol. 8, pp. 1-95, 01 1973. [4] T. I. Fossen and T. Perez, “Marine Systems Simulator (MSS),” https://github.com/cybergalactic/MSS, 2004, last accessed: 2022-06-14.

An image sequence dataset of growing snowflakes in HDF5 format. Generated by the Gravner-Griffeath LCA model for snow crystal growth. Useful for modeling crystal growth with neural networks.

Simulation data and pre-trained Graph Neural Network (GNN) models produced in [1]. Two .zip files are provided: data.zip - contains the datasets of train/test simulations produced using the Sheet Model algorithm [1, 2] models.zip - contains the GNN model weights (.pkl) + relevant training information and model parameters (*.yml and *.txt) Source Code The source code used to produce the data, train, and test the models can be found at: https://github.com/diogodcarvalho/gns-sheet-model References [1] D. D. Carvalho, D. R. Ferreira, L. O. Silva, "Learning the dynamics of a one-dimensional plasma model with graph neural networks", arXiv:2310.17646 (2023) [2] J. Dawson, "One‐Dimensional Plasma Model", The Physics of Fluids 5.4 (1962): 445-459.

Neural network model files and Madgraph event generator outputs used as inputs to the results presented in the paper "Learning to discover: expressive Gaussian mixture models for multi-dimensional simulation and parameter inference in the physical sciences" arXiv:2108.11481; 2022 Mach. Learn.: Sci. Technol. 3 015021 Code and model files can be found at: https://github.com/darrendavidprice/science-discovery/tree/master/expressive_gaussian_mixture_models

This repository contains data for a research project involving graph neural networks (GNNs) applied to mechanical metamaterials and their deformations. The mechanical metamaterials of interest consist of a flexible rubber-like material with a square grid of almost-circular holes in it, of various diameters. The holes are not quite circular, because they were made slightly elliptic to avoid the bifurcation point. This data was used to obtain the results in the paper 'Similarity Equivariant Graph Neural Networks for Homogenization of Metamaterials'. The data includes PyTorch Geometric Graph objects, raw data from MATLAB simulations, and files describing the finite element mesh. The Jupyter notebook 'data_create_graphs.ipynb' in the GitHub repository can be used to turn the raw data and mesh information into the graph objects. GitHub repository with the code: https://github.com/FHendriks11/SimEGNN ArXiv link to the corresponding paper Similarity Equivariant Graph Neural Networks for Homogenization of Metamaterials: https://arxiv.org/abs/2404.17365 The .pkl files are pickle files and can be opened in Python using the standard pickle library. The mesh files, which have the extension .mat, are MatLab files and can be opened either in MatLab or in Python using scipy.io.loadmat from the scipy library.

Workshop Tools Dataset Motivated by the need for a dataset that also includes inertial information about the objects, we contribute the following dataset. It contains 20 common workshop tools, and for each object: - a watertight triangular surface mesh; - a synthetic colored surface point-cloud; - ground truth inertial parameters; - ground truth part-level segmentation; and - a grasping reference frame. ## List of Objects - Adjustable Wrench - Bent Jaw Pliers - C Clamp - Electronic Caliper - Hacksaw - Machinist Hammer - Nut Screwdriver - Ruler - Socket Wrench - Vise Grip - Allen Key - Box Wrench - Clamp - File - Hammer - Measuring Tape - Pliers - Rubber Mallet - Screwdriver - Vise Clamp ## List of Files per Object Each object has its on dedicated folder containing the following files: - Frames.png : Picture of the point-cloud in point-cloud.ply with the mesh reference frames, the grasping reference frame and the centre of mass. The mesh reference frame is the frame all points are expressed relative to and the world "Origin" is written over its origin. The grasping reference frame was manually defined such as to express how a human worker would intuitively grasp the object. The centre of mass of the object is visualized as a purple ball. - Inertia.txt : Geometric and mass properties of the object as computed with the CAD software used to produce the triangular mesh. This file is not directly used by our software but can be more easily understood by humans. - Inertia.yaml : This file is read by our software to obtain the ground truth inertial properties as well as the transform that relates the grasping reference frame with respect to the mesh frame. - Materials.txt : Human-readable notes about the materials, and therefore the mass densities, used for the object parts. - mesh.ply : Binary millimiter-scaled triangular mesh with colored vertices and the following header: ply format binary_little_endian 1.0 comment SOLIDWORKS generated,length unit = millimeters element vertex 10913 property float x property float y property float z element face 21642 property uchar red property uchar green property uchar blue property uchar alpha property list uchar int vertex_indices end_header - point-cloud.ply : Colored and part-labeled point cloud with the following format header: ply format ascii 1.0 comment Created by Open3D element vertex 2000 property float32 x property float32 y property float32 z property float32 red property float32 green property float32 blue property uint8 segmentation end_header - reconstructed_mesh.ply : Triangular mesh reconstructed from the point-cloud in point-cloud.ply using the method described in the paper referenced below. Can be compared to the original mesh in mesh.ply to evaluate the quality of the reconstruction. ply format ascii 1.0 comment VCGLIB generated element vertex 256 property double x property double y property double z element face 500 property list uchar int vertex_indices end_header ## Citation If you used any part of this software in your work, please cite our paper: @inproceedings{Nadeau_PartSegForInertialIdent_2023, AUTHOR = {Philippe Nadeau AND Matthew Giamou AND Jonathan Kelly}, TITLE = { {The Sum of Its Parts: Visual Part Segmentation for Inertial Parameter Identification of Manipulated Objects} }, BOOKTITLE = {Proceedings of the {IEEE} International Conference on Robotics and Automation {(ICRA'23})}, YEAR = {2023}, ADDRESS = {London, UK}, MONTH = {May}, DOI = {} }