File size: 2,060 Bytes
5e00643 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 | # 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.
|