cryocare-v0.3.0-synthetic-parity

scitomo/cryocare-v0.3.0-synthetic-parity

cryocare-v0.3.0-synthetic-parity is a machine learning model from scitomo. Use it for the machine learning task on the model card, and read the license before you ship it in a product. It is set up for scitomo. The card lists the license as gpl-3.0.

This is a one-step analytic synthetic parity checkpoint, not a scientifically trained cryo-ET denoiser. FORMAT 2 closure does not validate it for cryo-ET scientific denoising, biological interpretation, Tomo110, or de…

Downloads · 30 days

Access

Public

Updated Sep 4, 2026

Repo size

8.2 MB

Likes

Public

Hugging Face

Repo makeup

Click a slice to open those files.

.safetensors4.1 MB · 97%

At a glance

Library: scitomo
License: gpl-3.0
Access: Public
Created: Aug 25, 2026
Updated: Sep 4, 2026
SHA: c41ada5f

Model card

From the Hugging Face model README

cryoCARE v0.3 synthetic parity checkpoint

It was freshly regenerated on CPU through cryoCARE 0.3.0 and converted from the new vendor archive only. Historical Hugging Face tensors, FORMAT 1 records, HDF5 archives, and old package artifacts were not conversion input.

FORMAT: closed learned-checkpoint FORMAT 2; manifest schema 4; revision 3.
Fresh vendor archive: 7559168 bytes, SHA-256 f483e3d3f771a1d8b68eb8e38e454f7912b8c0bb6d60130546f9b6e8fb20a8c2.
Fresh config.json: 596 bytes, SHA-256 5079346a1eb4dd40794b8b9b6b4497606367017fb56cb2c8ab0d05ab609c722b.
Fresh weights_best.h5: 4,157,904 bytes, SHA-256 f5f24cf72be041e206159896c99f22b5043b58631ba97707f12d215164acd3b0.
Fresh norm.json: 64 bytes, SHA-256 6bee6eba160161bb2c150a17f7adfd3975b38f05b25fb6f862c99592fd23f065; mean -0.03654887154698372, std 0.2582538425922394.
Scitomo authority: 731344180e7d8f7782ebc2ec0bd8a6aeeffa5f07 / 07b5fb8ed7292628b300115f1cc9f7df818a3d51.
cryoCARE_pip: 835e22c21031f967113502183413a0a9269c3c06 / d7c38f4e51e9634872711e2cc1cbff7bb686152c; cryoCARE_T2T: 17e95d8a17e31d5403c24aaaa8022a74974f06b9 / cef5e259f8a4aa093106aa99a94bd83deb988b7f.
Official cryoCARE-0.3.0.tar.gz source-distribution SHA-256: 8885aeb03d1731de1958463bbf766fa9999f4480d71c3af09f711ee210ab9886.
Vendor environment: Python 3.8.20, TensorFlow 2.4.4, Keras 2.4.0, CSBDeep 0.7.4, NumPy 1.19.5, h5py 2.10.0, float32 CPU; seed 20260825.
Network: 30 canonical tensors, 1,024,017 parameters.
Inference profile retains fitting mean/std and the proven depth-3 patch-support claim.
Every persisted MaxPool3D uses deterministic_nonoverlapping backward semantics.

conversion/conversion-record.json contains all 30 semantic mappings and exact source provenance. resources/conversion-run.py, evidence/forward-parity.json, and evidence/backward-parity.json record the fresh generic-loader closure. No HDF5, vendor archive, pickle, SavedModel, Keras object, NPZ, source checkout, or environment cache is published.

Library: scitomo
License: gpl-3.0
Created: Aug 25, 2026
Updated: Sep 4, 2026