Skip to content

Releases: DigitalHolography/EyeFlowMatlab

v3.0.0

Choose a tag to compare

@fischeryann fischeryann released this 09 Jun 14:06

v2.11.0

Choose a tag to compare

@fischeryann fischeryann released this 17 Apr 14:22
  • updated to fix incompatibilty errors with HoloDoppler 3.0.0
  • improved Folder Manager and GUI
  • export local segment values by default

Full Changelog: DigitalHolography/EyeFlow@2.10.1...2.11.0

v2.10.1

Choose a tag to compare

@fischeryann fischeryann released this 30 Jan 16:41

Full Changelog: DigitalHolography/EyeFlow@2.10.0...2.10.1

fixed a bug where the exe would

v2.10.0

Choose a tag to compare

@fischeryann fischeryann released this 29 Jan 14:36

v.2.9.11

Choose a tag to compare

@fischeryann fischeryann released this 15 Jan 13:48

Release v2.9.11

Date: 2026-01-15

Improvements

  • Synthetic spectral analysis
    • Simplified and cleaned function signatures
    • More consistent angle handling and velocity floor thresholding
    • Improved figure layout prior to export (aspect ratio, box, linewidths)
  • Pulse analysis
    • Corrected scaling factor
    • Added CorrectionCoeff parameter
    • Background filling switched to maskedAverage for improved stability
  • Cross-section & flow analysis
    • Removal of unused parameters
    • More robust handling of low-velocity and degenerate cases
  • Papilla detection
    • Improved handling of spatial reference and detection edge cases

Bug Fixes

  • Fixed frame capture logic in circleImages (now uses current axes for video resizing)
  • Fixed undefined variable access in createMask
  • Corrected method name typo: createtimeVector → createTimeVector
  • Fixed multiple output edge cases that could previously fail or return inconsistent results
  • Corrected background fill method inconsistencies (inpaintCoherent vs defaults)

Figures & Outputs

  • Improved figure aesthetics and export reliability
  • More consistent figure-saving pipeline
  • Cleaner handling of outputs in strokeAndTotalVolume
  • Removal of obsolete or redundant plotting paths

Maintenance & Cleanup

  • Removed deprecated ExtendedPulseAnalysis
  • Parameter cleanup across pulse, spectral, and cross-section analysis functions
  • Formatting and readability improvements in blood-flow velocity code
  • Improved warning and error messages

Summary

v2.9.11 improves numerical robustness, cleans analysis APIs, fixes multiple edge-case bugs, and significantly improves figure generation and output consistency.

What's Changed

Full Changelog: DigitalHolography/EyeFlow@2.9.10.2...2.9.11

v2.9.10.2

Choose a tag to compare

@fischeryann fischeryann released this 06 Jan 15:48
  • fixed an issue that called the getParams function instead of storing the parameters to avoid calls to outside memory

Full Changelog: DigitalHolography/EyeFlow@2.9.10.1...2.9.10.2

v2.9.10.1

Choose a tag to compare

@Zacharie-any Zacharie-any released this 16 Dec 16:57
  • latest

v2.9.10

v2.9.10 Pre-release
Pre-release

Choose a tag to compare

@Zacharie-any Zacharie-any released this 12 Dec 15:15

Full Changelog: DigitalHolography/EyeFlow@2.9.9...2.9.10

  • Should work with 10 workers under 128 GB of RAM
  • Considerable amount of changes including
  • Womersley number estimation for all segments
  • Velocity Histograms for all segments
  • Velocity distribution statistical outputs
  • Neural Net integration for better segmentation, diaphragm and papilla precise retrieval
  • eyeflow cli introduction for the incoming .exe version
  • In painting added for choroidal vessels effect compensation

v2.9.9

Choose a tag to compare

@fischeryann fischeryann released this 27 Nov 17:27

Warning: the 2.9.8 features are not inside this version a following patch will add them.

What's Changed

New Contributors

Full Changelog: DigitalHolography/EyeFlow@2.9.7...2.9.9

2.9.8

2.9.8 Pre-release
Pre-release

Choose a tag to compare

@noTban noTban released this 25 Nov 17:26
  • Add model versioning for artery/vein segmentation.
  • Add best segmentation model (https://arxiv.org/pdf/2511.14654)
  • For compatible environments (Python >3.10 & <3.13, Pytorch built with CUDA), use Pytorch models instead of ONNX.