| # Attribution and modification notice |
|
|
| ## Original material |
|
|
| **DeepDeWedge Tutorial Data** |
| Creator: Simon Wiedemann |
| DOI: <https://doi.org/10.6084/m9.figshare.25043435.v1> |
| Figshare file: `tutorial_data.zip`, file id `45582309` |
| Checkpoint member: `tutorial_data/fitted_model.ckpt` |
| License: Creative Commons Attribution 4.0 International |
| License URI: <https://creativecommons.org/licenses/by/4.0/> |
|
|
| The original checkpoint accompanies: |
|
|
| Simon Wiedemann and Reinhard Heckel, “A deep learning method for simultaneous |
| denoising and missing wedge reconstruction in cryogenic electron tomography,” |
| Nature Communications 15, 8255 (2024). |
| <https://doi.org/10.1038/s41467-024-51438-y> |
|
|
| The upstream DeepDeWedge implementation is available at |
| <https://github.com/MLI-lab/DeepDeWedge> and was reviewed at revision |
| `072075692a44a8f17394214369e6e762abe52bc3`. |
|
|
| ## Changes made by scitomo |
|
|
| The original PyTorch Lightning checkpoint was converted into a |
| scitomo-native package: |
|
|
| - executable/pickled training metadata was excluded; |
| - tensor state was exported in Safetensors format; |
| - 56 vendor state names were mapped through a reviewed explicit mapping to the |
| native scitomo `UNet3D` state; |
| - the vendor `unet.` namespace was removed; |
| - the second bottleneck convolution was mapped from vendor sequence index `2` |
| to the semantically equivalent native sequence index `4`; |
| - two learned normalization values changed storage role from non-trainable |
| parameters to native buffers without changing their values; and |
| - strict construction, inference, conversion, validation, provenance, and |
| tensor-inventory records were added. |
|
|
| No tensor value was intentionally changed. Synthetic forward output was |
| bit-exact, and a frozen real tutorial-volume crop passed the predetermined |
| relative-L2 parity threshold. |
|
|
| The converted package is distributed under the source material's CC BY 4.0 |
| terms. No endorsement by Simon Wiedemann, Reinhard Heckel, the Machine Learning |
| and Information Processing Laboratory, Nature Communications, or Figshare is |
| stated or implied. |
|
|
