Datasets are comprised of Image Collections and Snapshots. Image Collections are used to manage and annotate images, whereas a Snapshot is an Image Collection frozen at a single point in time used for model evaluation and training.
Only Snapshots can be exported, but Image Collections can be converted to Snapshots for export. To convert an Image Collection to a Snapshot, navigate to the Datasets tab on the left, select the Image Collection of interest, and then click on the “Generate Snapshot” button at the top.
Once you have a Snapshot ready to export, just navigate to it and click the “Download” button at the top.
The downloaded asset will be a zip file containing a COCO-formatted dataset, including the raw images and their annotations. Learn more about the COCO format.
Models trained with Fritz AI for mobile applications come in three different formats: TensorFlow Lite (for Android,
ModelName.tflite), Core ML (for iOS,
ModelName.mlmodel), and Keras (backend format,
ModelName.h5). All three formats are available for download. (Note: SnapML models are only available in a single format.)
To download a model, navigate to the Models tab on the left, select the model, and then click on the “Download” button next to the desired model version.
For SnapML projects, follow the same steps above to download either the raw model file or a working Lens Studio template project with the model baked in.