What's Changed

• Fix Error Handling and None Value Processing in Templates Aggregation Pipeline by @ShevchenkoRostyslav in #177
• Fix Error Handling and None Value Processing in Templates Aggregation Pipeline (#177) by @ShevchenkoRostyslav in #178

Full Changelog: v1.9.6...v1.9.7

Description

Quick fix that preserves an image_id within the aggregation pipeline.

What's Changed

• Preserve Image_id within the aggregation pipeline by @ShevchenkoRostyslav in #172
• Bump iris version by @ShevchenkoRostyslav in #174
• Release v1.9.6 by @ShevchenkoRostyslav in #175

Full Changelog: v1.9.5...v1.9.6

open-iris v1.9.5

Changed

• Add image denoising before segmentation
• Update the performance card

Release 1.9.4

Description

Replaced DistanceMatrix serialization for better reliability/portability and added broader unit tests for metadata JSON serialization to cover edge cases and ensure consistent round-trips.

Limitations

Previously serialized DistanceMatrix artifacts may require re-generation or a one-time migration.

What's Changed

• Release 1.9.4 by @ShevchenkoRostyslav in #158

Full Changelog: v1.9.3...v1.9.4

Release 1.9.2

Summary

• Add image_id to IRImage class #149

What's Changed

• Release 1.9.3 by @wiktorlazarski in #151

Full Changelog: v1.9.2...v1.9.3

Release 1.9.2

Summary

• Introduces a new interpolation fusion method for multiview/multiframe processing.
• Preserves numpy.ndarray in safe serialization (no implicit list conversion).

Key Changes

• New interpolation fusion method for multiview/multiframe pipelines.
• Serialization: keep numpy.ndarray unchanged; update tests to match.
• Version updates across codebase and pipeline configuration files.

Impact/Risk

• Potential adjustments where list conversion of arrays was assumed.
• Otherwise backward-compatible; no external API changes expected beyond serialization behavior.

What's Changed

• Release 1.9.2 by @ShevchenkoRostyslav in #147

Full Changelog: v1.9.1...v1.9.2

Hotfix: Preserve sequence types in __safe_serialize, add tests

Summary

• Preserve original sequence types (list vs tuple) during safe serialization.
• Add focused unit tests for __safe_serialize.
• Strengthen multiframe E2E checks.
• Fix documentation module reference.

Changes

• src/iris/orchestration/output_builders.py
  • _nested_safe_serialize: return type(obj)(...) for lists/tuples instead of always list[...].
  • __safe_serialize: return type(object)(...) for lists/tuples instead of always list[...].
• tests/unit_tests/orchestration/test_output_builders.py
  • Add TestSafeSerialize test suite (None, ImmutableModel, numpy array to list, list/tuple recursion with type preservation, primitive passthrough, unsupported type raises).
  • Access private serializer via getattr to avoid dunder name-mangling.
• tests/e2e_tests/pipelines/test_e2e_multiframe_iris_pipeline.py
  • Assert individual_frames[0]["metadata"]["image_size"] is a tuple.

Rationale

• Avoid unintended type coercion (tuple → list) during serialization, ensuring API consistency (e.g., image_size remains a tuple).
• Add regression coverage around serialization behavior.

Testing

• Unit: tests/unit_tests/orchestration/test_output_builders.py passes.
• E2E: strengthened assertion for multiframe pipeline metadata type.

Risk/Impact

• Low. Behavior change is limited to preserving container type during serialization; otherwise non-breaking.

What's Changed

• release 1.9.1 by @ShevchenkoRostyslav in #140

Full Changelog: v1.9.0...v1.9.1

Open IRIS v1.9.0 Release Notes

🚀 New Features

✨ Multiframe Iris Pipeline

• New MultiframeIrisPipeline class: A comprehensive pipeline that combines individual iris processing with template aggregation
• Unified configuration system: Single YAML configuration file that manages both iris processing and template aggregation phases
• Multi-image processing: Process multiple iris images and aggregate the resulting templates into a single high-quality template
• Enhanced output builders: New output builders specifically designed for multiframe pipeline results

🔧 Pipeline Architecture Improvements

• Renamed MultiframeAggregationPipeline to TemplatesAggregationPipeline: More descriptive naming for better clarity
• Enhanced output builders: Extended output_builders.py with new functions for multiframe pipeline outputs
• Improved error classes: Added new error types for better exception handling in template aggregation scenarios

🛠️ Infrastructure & DevOps

📚 Documentation Deployment

• Automated documentation deployment: New GitHub Actions workflow for automatic docs deployment on releases
• Enhanced build process: Improved documentation building with better handling of hidden files and build artifacts
• GitHub Pages integration: Streamlined deployment to GitHub Pages with proper .nojekyll file handling

📦 PyPI Publishing

• Automated PyPI publishing: New workflow for automatic wheel and source distribution publishing
• TestPyPI support: Support for publishing to TestPyPI for testing purposes
• Release-triggered publishing: Automatic publishing on GitHub releases and main branch pushes

🔍 Technical Details

Testing Improvements

• Comprehensive test coverage: Added extensive unit tests for the new multiframe pipeline (783+ lines of test code)
• End-to-end testing: New e2e tests for multiframe iris pipeline functionality
• Test refactoring: Improved existing tests to work with the new pipeline architecture

⚠️ Important Notes

Template Quality Requirements

The new multiframe pipeline includes important warnings about template quality and validation requirements. Users must ensure:

• Templates are from the same user and same eye
• Templates meet quality thresholds (no off-gaze, occlusion, poor focus)
• Appropriate validation criteria are established before production use

🔗 Migration Guide

For users upgrading from v1.8.x:

• The MultiframeAggregationPipeline has been renamed to TemplatesAggregationPipeline
• New MultiframeIrisPipeline class is available for combined processing
• Enhanced error handling provides better debugging capabilities
• Automated deployment workflows reduce manual intervention

What's Changed

• release v1.9.0 by @ShevchenkoRostyslav in #136

Full Changelog: v1.8.2...v1.9.0

open-iris v1.8.2

Changed

• Add option to load weights from a given path
• Bug fix on hamming distance with separate_half_matching enabled, the maskbit count should be doubled when applying HD normalization.

Bugfix

Removed the weights from the output_builders for the Multi-frame pipeline due to potential privacy concerns

What's Changed

• Iris multiview/aic 5848 by @ShevchenkoRostyslav in #116

Full Changelog: v1.8.0...v1.8.1