v0.6.1

Archive

This repository will now be archived as I plan to support other projects.

This last release update requirements to avoid having dependency issues.

What's Changed

• [pre-commit.ci] pre-commit autoupdate by @pre-commit-ci in #65
• [pre-commit.ci] pre-commit autoupdate by @pre-commit-ci in #66
• [pre-commit.ci] pre-commit autoupdate by @pre-commit-ci in #67
• [pre-commit.ci] pre-commit autoupdate by @pre-commit-ci in #68
• [pre-commit.ci] pre-commit autoupdate by @pre-commit-ci in #69
• [pre-commit.ci] pre-commit autoupdate by @pre-commit-ci in #70
• Update requirements.txt by @juliendenize in #72
• [pre-commit.ci] pre-commit autoupdate by @pre-commit-ci in #71

Full Changelog: v0.6.0...v0.6.1

v0.6.0

What's Changed

Release of nested tensors

In this release, we provide a new set of tensors called nested tensors that allow to transform tensors of different shapes.
It does not rely on the beta tensors from Torch but if they are to be out at some point it will be the case.

Support of Python 3.12

With the new release of Torch 2.4, there is now the possibility to support python 3.12 for Torchaug !

Other

• to_image now supports pillow images
• the license header is now checked via the CI.

Commits

• Add nested tensor by @juliendenize in #58
• [pre-commit.ci] pre-commit autoupdate by @pre-commit-ci in #57
• [pre-commit.ci] pre-commit autoupdate by @pre-commit-ci in #59
• Feat/read pillow by @juliendenize in #60
• Add doc for nested tensors by @juliendenize in #62
• Ci/header by @juliendenize in #63
• Add python 3.12 support by @juliendenize in #64

Full Changelog: v0.5.2...v0.6.0

v0.5.2

What's Changed

• Add torchvision version test by @juliendenize in #55

Full Changelog: v0.5.1...v0.5.2

v0.5.1

What's Changed

• Fix getter of batch size and device for flat inputs in transforms by @juliendenize in #50
• Update requirements to use Torchvision 0.18.1 by @juliendenize in #51
• Fix using pin_memory with Torchaug tensors by @juliendenize in #52

Full Changelog: v0.5.0...v0.5.1

v0.5.0

Highlights

Add the JPEG and RGB transforms introduced by Torchvision.

What's Changed

• Fixed wrong link by @vfdev-5 in #30
• [pre-commit.ci] pre-commit autoupdate by @pre-commit-ci in #33
• [pre-commit.ci] pre-commit autoupdate by @pre-commit-ci in #34
• [pre-commit.ci] pre-commit autoupdate by @pre-commit-ci in #35
• [pre-commit.ci] pre-commit autoupdate by @pre-commit-ci in #36
• [pre-commit.ci] pre-commit autoupdate by @pre-commit-ci in #37
• [pre-commit.ci] pre-commit autoupdate by @pre-commit-ci in #38
• Add JPEG transform by @juliendenize in #40
• Add RGB transform by @juliendenize in #41
• Improve SanitizedBoundingBox by @juliendenize in #42

New Contributors

• @vfdev-5 made their first contribution in #30

Full Changelog: v0.4.2...v0.5.0

v0.4.2

Highlights

• Add labels ta tensor by @juliendenize in #24
• Fix chunk indices by @juliendenize in #26
• Fix transform with chunks by @juliendenize in #27
• Add coco example by @juliendenize in #29

Full Changelog: v0.4.1...v0.4.2

v0.4.1

Highlights

Torchaug now follows Torchvision v2 transforms:

• TA Tensor or Torchaug tensors behave similarly to Torchvision's TV Tensor with new types to handle batches of data.
• transforms are only located in one package: torchaug.transforms and torchaug.batch_transforms is no longer supported.
• All transforms in Torchvision v2 are also accessible in Torchaug.
• LICENSE is now CeCILL-C.

Full Changelog: v0.3.3...v0.4.1

v0.3.3

Highlights

• Add RandomGrayscale #14
• Refractor code and docs to match Torchvision transforms category #13

Other

• Rename BatchRandomGrayScale to BatchRandomGrayscale #14

v0.3.2

Highlights

• Add VideoBase #12
• Fix inplace error when batch random transforms have p=1 inside a wrapper d9ad37b
• Fix video_format override by wrappers in video transforms inside the wrapper #12

Other

• Add log api in image wrappers
• Fix some of the doc

v0.3.1 Hotfix

Highlights

• Hotfix v0.3.0 Wrapper that handled transforms wrongly
• Improve documentation for wrappers
• Set inplace default parameter to False for all BatchRandomTransform subclasses as wrappers should handle them for augmentations pipelines.