Arnas Scope 1.2.0 Release Notes
Released:
Arnas Scope 1.2.0 adds a settings recommender, new regularization options, confocal point-spread function (PSF) models, self-service activation, and a round of accuracy fixes. It also lets very large datasets finish processing. The fixes from 1.1.4 are still in place. Some results will look different from 1.1.4. See “What you will see change” below.
What's new
- Command-line tool included. The Arnas Scope command-line tool now installs with the desktop app. Your Arnas Scope license covers both.
- Settings recommender. The Autotune button examines the loaded volume (camera gain, offset, background noise, and signal-to-noise) and suggests starting settings: iterations, regularization weight and mode, background handling, guard band, Z super-resolution, and padding. You can keep a local history of your runs, and the closest earlier recipes appear next to each recommendation with their results. A starter set of recipes is included. The recommender doesn't send your data anywhere.
- More regularization choices for 3-D Richardson-Lucy: RL-OMW, TV-ADMM, FISTA-TV, and direct filters (Tikhonov, Wiener, regularized inverse), next to the existing TV-style curvature regularizer. The TV-style regularizer now works over an effective weight range of about 0.003 to 0.2. Higher values level off instead of making the run unstable.
- Confocal PSFs. Adds line-confocal and rescan-confocal PSF models. Blind Zernike PSF estimation adds an astigmatism option.
- More control over noise and the camera: gain-aware damping, a camera defect map input, an axial noise estimator, and a guard band that survives 16-bit export and re-import.
- Clearer background options: none (no offset is estimated or subtracted), manual, and auto (estimated from the darkest corner).
- Two PSF files. Each PSF is saved as a full-precision file for measurement and reuse, plus a 16-bit copy for viewing. The four-pane comparison uses the full-precision file.
- Limit on blind PSF spread. Blind PSF estimation can keep at most 5% of the PSF's energy outside the starting PSF's footprint (adjustable). Diagnostics report how much was used.
- Self-service activation. The activation dialog shows your Machine ID and opens https://arnastech.com/license, so you no longer need to email us to activate. Contact: [email protected].
- Choose where scratch files go. Streamed deconvolution can write its temporary volume to a drive you pick.
Changes
- New recommended settings. The defaults now use Richardson-Lucy with the TV-style regularizer.
- Light-sheet: conservative (15 iterations, weight 0.1, with intensity enforcement), standard (25 iterations, 0.05), and aggressive (40 iterations, 0.02).
- Confocal: RL-TV at 20, 30, or 60 iterations, weight 1e-4.
- RL-OMW and TV-ADMM are still available as faster presets. When the recommender can estimate better values from your data, it uses them.
- Tile overlap now follows the PSF. The overlap between tiles grows to the PSF's half-width, up to 64 voxels or a quarter of the tile size. The command-line tool logs the value it chose.
- Better memory protection when building PSFs. Arnas Scope estimates the memory needed before building a PSF. If memory is tight, the app asks you to proceed or cancel before the run starts. If the system still can't allocate the memory, the PSF is built at slightly lower detail and you get a warning, rather than the run failing. Blind PSF estimation now has the same protection.
Fixes
- Very large streamed datasets now finish. In our testing, memory use stays flat at about 2.8 GB. Before, it grew from about 5 GB to over 26 GB and could stop partway through datasets of several hundred GB.
- Background was subtracted twice on some processing paths. Now it's subtracted once.
- The guard band was redesigned. The saturation threshold is now explicit, and the band works in the same units as the processing.
- Theoretical and measured PSFs are now averaged over the detector pixel on every processing path, not just the two main ones.
- Intensity enforcement no longer brightens dim volumes by the camera offset.
- Negative regularization weights are rejected.
- Choosing the “TV-style curvature regularizer” in the app could make the run fail. Fixed.
What you will see change: dim data with a high camera offset now comes out with the right total intensity, saturated regions inside the guard band no longer ring, and the GPU and alternate processing paths now give matching results.
License
The license is updated to version 2026-10-09. You'll see it during installation, and it covers both the desktop app and the command-line tool. You can still activate by placing your arnasscope.lic file in %APPDATA%\ArnasScope, or by redeeming a code at https://arnastech.com/license. If your older license file says “professional” or “academic,” it counts as a paid Standard seat and allows commercial use, unless a separate written agreement says otherwise.
Upgrading from 1.1.4
Run the 1.2.0 installer over your existing copy. Your settings are kept.
System requirements
- 64-bit Windows
- An NVIDIA GPU with a recent NVIDIA driver, version 566.36 or later
- The CUDA 12 runtime comes with Arnas Scope, so you don't need to install the CUDA Toolkit.
- Installer download: about 1.1 GB