We have a new open data set, higher statistics and higher spatial resolution samples, hosted at Open Science Framework (OSF)! Using the OSF is a new attempt but we like it for several reasons: it can generate DOI for citations, recognized by popular publishers (like Nature), comes with command-line-interface (for downloading/uploading data), and most importantly it's free! Another data hosting tier we could use is HEPData, possibly in future.
If you kindly consider using our data, please cite DOI 10.17605/OSF.IO/VRUZP and the credits go to DeepLearnPhysics group as a whole.
The data samples are the simulation of liquid argon time projection chambers (LArTPCs) with 3mm/pixl resolution generated in three different cubic volumes: 192, 512, and 768 pixels (the side length of a cube). The samples are shared in both 3D and 2D projections (XY, YZ, ZX planes). Information included are energy depositions of generated particles, pixel-level segmentation labels, particle-level pixel cluster information, and other particle-wise information such as energy, particle type, start/end points, etc.. One can use this sample for semantic segmentation, object detection, particle clustering, particle type classification, as well as energy regressions.
The data is used for the first demonstration of sparse submanifold convolutional neural network (SSCN) for scalable Deep Learning techniques for liquid argon time projection chambers. You can read the paper. There are a few projects undergoing within the group to utilize this data for studying/demonstrating other ML techniques. Stay tuned!