- moving data between Firestore and Labelbox
- import and export labels from Labelbox
- importing predictions to Labelbox via API
- exporting annotations from Labelbox
- Install go version 1.13 from https://golang.org
- Generate service account key for Firestore and place it in
./.secrets/service-account.json
- Generate secret API token for Labelbox and place it in
./.secrets/labelbox-api-token.txt
- (optional) install
make
if you want to use commands from Makefile
cmd/cli
- get data from Firestore
- transform Open Images labels to format needed for Labelbox import
- transform Open Images annotations to format needed by our evaluation scripts
cmd/compute-dataset-stats
- compute stats about collected datasetcmd/construct-label-mapping
- debug reconstruction of labelscmd/download-openimages-test-images
- download subset of test images from Open Imagescmd/export-dataset-datarows
- export rows from Labelboxcmd/format-openiamges-test-images-annotations
- format GT boxes provided by Open Images into files with format needed for our evaluation scriptcmd/init-prediction-model
- initialize prediction model in Labelbox using APIcmd/make-prediction-model
- import predictions made by models into Labelbox using APIcmd/switch-prediction-model
- switch active prediction model in Labelbox using API
Note: you need to have secrets in place to run these programs.
- Run
make run-make-labelbox-labels
from the root to transform labels from Open Images to JSON file that can be imported to Labelbox when creating new project for annotating. - Run
make run-labelbox-annotations-to-validation
to transform manually exported annotations from Labelbox to format that is needed for our scripts. - Run
make run-get-images
to download annotated images from Firestore. - Run
go run cmd/export-firestore-dataset/main.go
to export collected dataset from Firestore to /data/collected-dataset.
Every program in cmd
directory can be ran as go run cmd/<name_of_the_program>
from the root directory.