CLI Tools

Caipy to Fiftyone

Directly convert a Caipy folder into fiftyone for quick inspection

usage: caipy_to_fiftyone or c2f [-h] [--mode {vanilla,generic}]
                                [--input-folder INPUT_FOLDER]
                                [--images_root IMAGES_ROOT]
                                [--annotations_root ANNOTATIONS_ROOT]
                                [--splits-to-read [SPLITS_TO_READ ...]]
                                [--dataset-name DATASET_NAME]
                                [--not-persistent] [--no-app] [--watch]
                                [--port PORT] [--only-local]

Input arguments

--mode

Possible choices: vanilla, generic

Choice between vanilla mode, where a single folder is given with option --input_folder, and generic, where two folders are given with options --images_root and --annotations_root

Default: 'vanilla'

--input-folder, -i

When in vanilla mode, folder where the CAIPY dataset is stored

--images_root, --ir

When in generic mode, folder where images are stored. Equivalent folder in vanilla mode is dataset/Images

--annotations_root, --ar

When in generic mode, folder where annotations json files are stored. Equivalent folder in vanilla mode is dataset/Annotations

--splits-to-read, -s

Optional list of splits to read. If not selected, will read all splits and convert them to fiftyone

Fiftyone arguments

--dataset-name, --name, -n

Optional dataset name to appear in fiftyone app. If not selected, will take the name of loaded dataset object, i.e. the name of input-folder or parent of --images-root without its parents

--not-persistent, --np

If selected, will not save the converted dataset in fiftyone. It will be available to inspect with the browser during the time this script isrunning, but will be removed as soon as the script is stopped.

--no-app, --na

If selected, will not launch the app. This needs the ‘persistent’ option to be selected for the command to do anything

--watch, -w

If selected, will watch the folder containing the dataset. Each time a change in the folder is detected, the dataset is updated and so is the fiftyone counterpart. Useful when you are constructing a dataset gradually. Note that this option is useless if the app is not launched

--port, -p

Server port to connect to the app server. If not set, will choose randomly a free port

Default: 0

--only-local

if selected, will not open the server for other than localhost. the server can still be reachable with ssh tunnelling though