This code is for the paper "Autoregressive Models for Crystallographic Orientation Prediction in the Presence of Symmetries."
Paper can be found here
Most of the code is inspired by this repository. The EulerNet results can be viewed there.
Run main.py to obtain the following results of the SequenceNet. Test Time augmentation results are also noted below. The model and the data will be downloaded automatically.
| Metric | Value |
|---|---|
| Test misorientation median | 4.74 |
| Test misorientation mean | 8.24 |
| TTA misorientation median | 3.96 |
| TTA misorientation mean | 6.70 |
Run train.py to train the sequence net model with the default settings.
Using CUDA with a GPU is recommended.
Please download the data folder (3.5 GB) from the Mendeley Dataset available at DOI:10.17632/z8bh7n5b7d.1. Copy the data folder into the root directory of the repository.
Or run main.py to download the data.
The data folder contains:
(i) All training and evaluation sets used to derive the results.
(ii) Three additional files:
- /samples/08/drm_data.npy: A 4D numerical matrix (shape (x, y, theta, phi), type uint8) representing the experimental DRM dataset of the test specimen showcased in Figure 3 of the paper.
- /samples/08/eulers.npy: The corresponding matrix of Euler angles measured by EBSD for this test specimen (shape (x, y, 3), type float32).
- anomaly_specimen.npy: The DRM dataset of the specimen shown in Figure 6 of the paper to demonstrate the detection of out-of-distribution data.