Downloads · 30 days
0
scitomo/topaz-v0.2.4-10a
topaz-v0.2.4-10a 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. The card lists the license as gpl-3.0.
This is a Scitomo FORMAT-2 conversion of the official Topaz v0.2.4 pretrained 3-D 10a denoiser.
Downloads · 30 days
0
Access
Public
Updated Sep 5, 2026
Repo size
11.7 MB
Likes
0
Public
Click a slice to open those files.
.safetensors11.7 MB · 99%
From the Hugging Face model README
This is a Scitomo FORMAT-2 conversion of the official Topaz v0.2.4 pretrained 3-D 10a denoiser.
Upstream source: tbepler/topaz, revision c6dde54398875dcc6a210f83de019c9165fb474c, tree 25396d246daab8de75fb1adce57f7ad5cc883bf4.
The authoritative PyPI source distribution SHA-256 is e59200127fda2c23f5357c86e2c5f3e4ddfd5c103cf0b7138e700cac5b1fc516. The matching vendor checkpoint SHA-256 is 9011778cc231d667146435eba241fa984783624eaf18b929a0eabf23df527c3a.
The package has 34 canonical tensors / 2,916,337 parameters. Dense native parity was bitwise exact before export and after ordinary generic FORMAT-2 reload in the frozen conversion environment.
Whole-volume population z-score and output mean/std restoration are Topaz inference-profile semantics. Exact local valid-core support is 160 samples each side with modulo-32 core/origin/evaluation requirements, while the dense Network itself requires only spatial axes >=32 and no modulo-32 divisibility.
The historical vendor tiled patch_size=96, padding=48 path is distinct evidence and is not exact execution of the dense operator.
The upstream Topaz GPLv3 license text is retained at resources/TOPAZ_LICENSE.txt.
Redistribution of the bundled v0.2.4 checkpoint is approved under that upstream GPLv3 package license. This FORMAT-2 package revision 2 supersedes the initial immutable publication a688cc4f217f32d449ffcde2df272e05018a2ae5, whose manifest conservatively recorded redistribution as unknown. The learned weights, Network construction, inference profile, and scientific validation are unchanged.
No retraining or new scientific-quality claim is made by this conversion.