Releases: v7labs/darwin-py
Releases · v7labs/darwin-py
v0.6.2 Released
- Fixes issues with
:
_
and-
in dataset export names. - Fixes issues with floats in pascal voc files
- Detects if darwin-py version is out of date and warns.
- Fixes issues with downloading images with invalid names for windows.
v0.6.1 Released
Fix: allow underscore _
characters in slugs
v0.6.0 Released
New Upload API
We supercharged our upload API, so it's much easier to do more with your data at upload time.
Specify remote paths, tags, and other metadata for each individual uploading file
# Define each individual uploading file, with their own paths and tags
animal_files = [
LocalFile(DATA_DIRECTORY / "cat_1.jpg", path="cats", tags=["shorthair"]),
LocalFile(DATA_DIRECTORY / "cat_2.jpg", path="cats", tags=["ragdoll"]),
LocalFile(DATA_DIRECTORY / "cat_3.jpg", path="cats", tags=["persian"]),
LocalFile(DATA_DIRECTORY / "cat_4.jpg", path="cats", tags=["sphynx"]),
LocalFile(DATA_DIRECTORY / "cat_5.jpg", path="cats", tags=["unknown"]),
LocalFile(DATA_DIRECTORY / "dog_1.jpg", path="dogs", tags=["labrador"]),
LocalFile(DATA_DIRECTORY / "dog_2.jpg", path="dogs", tags=["unknown"]),
LocalFile(DATA_DIRECTORY / "dog_3.jpg", path="dogs", tags=["german-shepherd"]),
LocalFile(DATA_DIRECTORY / "dog_4.jpg", path="dogs", tags=["beagle"]),
LocalFile(DATA_DIRECTORY / "dog_5.jpg", path="dogs", tags=["bulldog"])
]
# Use your RemoteDataset object to push those files
dataset.push(animal_files)
Upload entire folders by keeping the tree structure intact on Darwin
dataset.push("/path/to/folder", preserve_folders=True)
The same can be done via the CLI!
darwin dataset push team/dataset /path/to/folder --preserve-folders
Only upload files if they're not already represented in the Dataset
At upload time, you will be warned if some of your files have been previously uploaded in the specified path.
$ ~/Downloads darwin dataset push team/dataset /path/to/images
Total progress ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 23 of 23
Skipped 2 files already in the dataset.
Sort files when listing them
We made it easy to sort files by inserted_at
, updated_at
, file_size
, filename
and priority
. Of course, you can specify if you'd like to sort in ascending or descending order, all in the same argument.
dataset.fetch_remote_files(sort="priority:desc")
The same can be done via the CLI!
darwin dataset files team/dataset --sort-by priority:desc
Note that the default sorting argument is set to updated_at:desc
.
Miscellaneous
- We now use Rich to render every message, progress bar, error or table in the CLI
- Specify callbacks to be called at upload time to manage progress
- Get clear error messages if the uploaded files fail at any stage of the upload process
v0.5.20 Released
- fixed issue with file name matching on imports
v0.5.19 Released
- Remove mandatory dependency on scikit-learn
- Make it possible to split video annotations into single frame annotations
v0.5.18 Released
- use
--append
to append (instead of overwriting) annotations during import - fixed issue where width and height got swapped for coco exports.
- updated
fetch_items
to handle more and longer filters.
v0.5.17 released
- Support for binary coco RLE
- Improved error message parsing
- Using a new version of upolygon to handle issues with very large polygons.
v0.5.16 released
- Add .TIFF support
- Fix importing annotations with attributes
- New archive/restore payload structure
- Avoid stripping class names when reading class list
- Fix symlink unlink on Windows
- Remove WORKFLOW feature flag checks
v0.5.15 released
- Speed up of annotations importing
- Support for .svs uploads
- Fixed bug in cross contamination
- Fixed issue in exporting tags for videos
- Use upolygon’s RLE ( ~15 times faster)
- Function for computing annotation density
- Optional rounding for COCO exports
- Use png for encoding video frame downloads
- Handle picture mode 1
v0.5.14 released
- Bug fix for importing complex RLE masks from coco.
- Removes extra prints during imports.
- Updated readme with links to documentation.