Releases: orobix/quadra
Releases · orobix/quadra
v1.1.1
What's Changed
Full Changelog: v1.1.0...v1.1.1
v1.1.0
What's Changed
- add: Migrate vit gradcam code + pytorch export format for classificat… by @AlessandroPolidori in #7
- Fix/binary linear model wrapper fix by @AlessandroPolidori in #12
- Bug: non squared input images break SklearnClassification task by @AlessandroPolidori in #15
- Docs/add newformat classif tutorial by @AlessandroPolidori in #18
- Fix/print config in anomaly base by @AlessandroPolidori in #19
- add: Add patch classification to tasks.md by @AlessandroPolidori in #24
- Feature/landing page by @rcmalli in #23
- Refactor/evaluation task by @AlessandroPolidori in #22
- Feature/experiment manager by @lorenzomammana in #3
- Feature/documentation Github Actions by @rcmalli in #29
- Fix/network builder kwargs by @lorenzomammana in #30
- fix: update readme links to always latest by @rcmalli in #32
- fix: update image links with correct relative path by @rcmalli in #33
Full Changelog:
Added
- Add ModelManager class to manage model deployments on model tracking platforms (e.g. MLFlow).
- Add automatic storage of file hashes in BaseDataModule class for better experiment reproducibility and tracking.
- Automatically load transforms parameters from model info file in base Evaluation task.
- Add support for vit explainability using Attention Gradient Rollout method.
- Add export of pytorch model for Classification and SklearnClassification tasks.
- Add automatical detection of model input shapes for better exportation capabilities. Add support for custom input shapes like models with multiple inputs.
- Add documentation landing page, improve colore themes and logo.
- Add github actions to automatically build and deploy documentation of main and dev PRs.
Changed
- Refactor evaluation task to be more generic and improve inheritance capabilities.
- Refactor export_types parameter to export for better configurability of export parameters.
- Change input_size model info parameter from HXW to a list of actual model parameters for inference (e.g [(3, 224, 224)]).
Fixed
- Fix gradcam not working properly with non rectangular images.
- Fix logistic regression wrapper not working properly with 2 classes for torch classification.
- Fix wrong typings in NetworkBuilder's init method.
- Fix broken links in documentation.
- Fix minor documentations issues.
v1.0.3
What's Changed
Fix circular import making some tests not running properly by @lorenzomammana in #28
v1.0.2
What's Changed
Fixed anomaly detection training not working with non convential image extensions (e.g. .BMP)
Hotfix segmentation datamodule
This release aims to target an error on the segmentation datamodules for which the validation dataset was used in place of the train dataset.