- Upgraded wai.annotations.audio to 1.0.2
- Upgraded wai.annotations.core to 0.2.1
- Upgraded wai.annotations.yolo to 1.0.2
- Upgraded wai.common to 0.0.42
- wai.annotations.tf is now optional
- install.sh now has -o flag to install optional modules when installing latest (-l)
- Upgraded wai.annotations.commonvoice to 1.0.2
- Upgraded wai.annotations.core to 0.2.0
- Upgraded wai.annotations.imgaug to 1.0.6
- Upgraded wai.annotations.layersegments to 1.0.2
- Upgraded wai.annotations.video to 1.0.2
- Upgraded wai.annotations.yolo to 1.0.1
- Added wai.annotations.coqui 1.0.0
- Added wai.annotations.audio
- Added wai.annotations.generic
- Upgraded wai.annotations.core to 0.1.8
- Upgraded wai.annotations.subdir to 1.0.1
- Added wai.annotations.grayscale
- Upgraded wai.annotations.bluechannel to 1.0.2
- Upgraded wai.annotations.coco to 1.0.2
- Upgraded wai.annotations.core to 0.1.7
- Upgraded wai.annotations.grayscale to 1.0.1
- Upgraded wai.annotations.imgaug to 1.0.5
- Upgraded wai.annotations.imgstats to 1.0.3
- Upgraded wai.annotations.imgvis to 1.0.3
- Upgraded wai.annotations.indexedpng to 1.0.1
- Upgraded wai.annotations.redis.predictions to 1.0.2
- Upgraded wai.annotations.roi to 1.0.1
- Upgraded wai.annotations.video to 1.0.1
- Added wai.annotations.imgstats.
- Added wai.annotations.imgvis.
- Added wai.annotations.opex.
- Added wai.annotations.video.
- Added wai.annotations.redis.predictions.
- Upgraded wai.annotations.core version.
- Upgraded wai.annotations.tf version.
- Included wai.annotations.imgaug.
- Included wai.annotations.yolo.
- Upgraded wai.annotations.tf version from 1.1.0 to 1.2.0.
- Included wai.annotations.voc.
- Upgraded wai.annotations.tf version from 1.0.0 to 1.1.0.
- Fixed build (hopefully).
- Split packages into separate modules so individual formats can be used/excluded.
- Sub-commands (convert, plugins, domains).
- Can use --help/-h anywhere in a command string and get help.
- Updated documentation.
- Bug fixes.
- Bug fixes.
- Added ability to provide a source iterable when using iterate_files.
- ConversionPipelineBuilder now works with any combination of sources, processors and sinks.
- Data elements now are only referred to by their base filename.
- Added the layer-segments image-segmentation format.
- Added the od-to-is cross-domain converter.
- Bug fixes.
- Fixed bug in calculating output directory from command-line option.
- Added strip-annotations ISP.
- Stopped COCO/ROI/VGG image object-detection formats from writing negative annotations.
- Fixed bug in coercions where negatives instances weren't being handled correctly.
- Added --dense option to to-tf-od which encodes masks in the dense numerical format.
- Refactor of the internal structure to aid in debugging.
- All input/output formats now end in the short-code of their domain.
- Fixed bug where --category-output-file was being written to an arbitrary location.
- Added --category-output-file option to to-coco format for writing the categories to a simple comma-separated text file.
- Added the subdir image classification format.
- Added the ADAMS image classification format.
- Refactored label-mapping, image format conversion and label filtering into their own ISPs (were options on object detection readers/writers).
- Better error messages when domains are incorrectly matched.
- Upgraded wai.common requirement to fix bug with file iteration.
- Made help more helpful.
- Added macros, which allow for simple command-line argument substitution.
- Improved logging during split-writes. More improvements to come.
- Separated base reader/writer functionality into abstract bases which don't assume local disk access.
- Added new option to the MS-COCO format to sort the categories.
- Added new options to the MS-COCO format for pre-specifying the categories to expect.
- Added domain information to the --list-plugins command.
- Basic support for speech annotations, including CommonVoice and FestVox formats.
- Added --debug global flag to aid in debugging.
- Added file_iterator method to conversion chains, allowing iteration over the converted file-data.
- Fixed dependency error on installation from PyPI caused by pycocotools.
- Bug fixes.
- Major restructure to introduce domains other than object-detection in images.
- Modified numpy/pycocotools requirements to work with EfficientDet.
- Added sha256, is_crowd and area features to TFRecords format to work with EfficientDet.
- Fixed bug where full set of TFRecord shard files wasn't being generated.
- Added InputChain and OutputChain classes, encapsulating the stages of input and output respectively.
- Modified parsing so that users can create "main-like" CLI interfaces.
- Bug fixes.
- Updated requirement for wai.common to v0.0.31.
- Updated requirement for wai.bynning to v0.0.2.
- Added support for BMP-format images.
- Added
--convert-image
option to convert images to a given image-format.
- Added
--split-names
and--split-ratios
options for creating splits in the dataset. - Added
--seed
option for randomising file-read order.
- Bug fix where plugin registry was being recreated every run.
- Bug fix where previous registry incarnation was being loaded instead of new version.
- Converted formats to use a plug-in system so other libraries can define their own formats.
- Upgraded wai.common requirement to v0.0.28.
- Separated main settings from library settings.
- Changed
--include-zero-area
flag to min/max width/height/area flags. - Added
--comments
option for ROI writer to insert comments at the beginning of written files. - Input directories read all contained files ending in
.report
when usingadams
format. - Added
--sample-stride
option to tfrecords input which sub-samples the mask when generating the polygon for speed increase. - Added
--mask-threshold
option to tfrecords input which sets the probability to consider the mask inside the polygon. - Internal format is now an object rather than a tuple.
- Added
--size-mode
flag to ROI format which writes ROI files with x,y,w,h headers rather than x0,y0,x1,y1 headers.
- Upgraded wai.common requirement to v0.0.26.
- VGG/COCO formats can now take a
--pretty
flag to pretty-print their JSON annotations. - Upgraded wai.common requirement to v0.0.25.
- image_utils.lists_to_polygon method converts lists of X and Y coordinates into a planar.Polygon
- image_utils.polygon_to_bbox method returns the x0,y0,x1,y1 coordinates of the bounding box around the supplied planar.Polygon object
- Conversions now automatically discard annotations with zero area, can be reverted
with
--include-zero-area
flag. - image_utils.mask_to_polygon can now work with a view of the mask instead of the full mask to speed up the polygon detection
- Modified utilities for getting labels/prefixes to be able to specify a default value.
- Upgraded wai.common requirement to v0.0.24.
- Added requirement for wai.json v0.0.4.
- Added compatibility layer for Tensorflow V1.
- Now ensures Tensorflow is in eager execution mode.
- Upgraded wai.common requirement to v0.0.22.
- Fixed default filename for ROI format ("-rois.csv" rather than "-roi.csv").
- Added the ability to specify a prefix and suffix for reading/writing ROI files.
- Logging changes.
- Moved utilities for each format into their own sub-package.
- Added reader options to read input files from list files.
- Separated command-line option parsing from io/converter classes so that imports unique to specific formats are not required unless using that format.
- Refactoring of package structure.
- Added logging of when files are being written.
- Conversion to COCO format now report polygon mask area, rather than just bounding- box area.
- Added support for non-standard keywords to the ROI format.
- Added the ability to specify inputs by glob syntax, and also specify negative images with no annotations.
- Added utility functions for images (image_to_numpyarray, remove_alpha_channel) and for masks (mask_to_polygon, polygon_to_minrect, polygon_to_lists) in module wai.annotations.image_utils
- Added --extensions option to control the search order for associated images.
- Added --verbosity flag to control logging verbosity.
- Added polygon mask support to the ROI format.
- Modified requirements to include Cython as this is required in later versions of pycocotools. Moved pycocotools to end of dependency list to ensure all requirements are met before installation.
- Changed --coerce option to --force.
- Tensorflow is no longer a dependency in setup.py as can work with tensorflow or tensorflow-gpu.
- TFRecords format can now handle polygon-mask annotations.
- Can now work with images that contain no annotations.
- Initial release