SiDIT

This repository contains tensor-only parameters for the paper's SiDIT base model, which jointly predicts reflection families and crystal properties from single-crystal diffraction point clouds.

The inference implementation and numerical preprocessing are available at max6616/SiDIT on GitHub. Recovered test inputs are distributed in max6616/SiD3M-test; consult the dataset card for their exact scope.

Parameters and configuration

sidit-base-state-dict.pt contains 523 state tensors, 194,915,607 bytes in total. The file contains the model parameters used for the benchmark and excludes optimizer state and Python argument objects. Loading uses torch.load(path, map_location='cpu', weights_only=True); construct the model with the accompanying config.json and the GitHub implementation.

• Backbone: Point Transformer V3 encoder–decoder, base configuration.
• Per-point classes: 21 classes for each of h, k, and l, corresponding to [-10, 10].
• Global outputs: crystal system, Laue class, point group, space group, and the nine-component cell head.
• Cell head: normalized lengths plus cosine/sine angles; decode in float32 as documented in the implementation.
• The checkpoint was saved at global step 80,000.

Model-file SHA-256: 22fceed0e9e303e97ce668cf71b092f39ab9446fe38ec350a4189bc97e389ad7.

Scope

The model produces canonical reflection-family labels and cell/symmetry candidates from the supplied simulated point-cloud format. Signed indexing additionally requires a common orientation and geometric verification and refinement. The dataset card specifies raw-input availability and benchmark result support.

Authors: Zhao Zhang, Zheng Dong, Zhi Geng, Xinlong Dong, Yi Zhang, and Gaoqi He. Original implementation copyright 2025 Zhang Zhao; MIT license. Pointcept-derived code retains its upstream MIT notice in the software repository.