Importing, exporting, and downloading PowerAI Vision information

You can import and export PowerAI Vision models and data sets. This allows you to save them for archiving then use them later, use them on a different PowerAI Vision install, and so on.

Exporting

Export a data set
To export a data set, open the Data sets page, open the data set you want to export, then click Export in the action bar. The data set is saved in your default download directory as data_set_name.zip. This zip file contains the images as well as any tags or categories you have assigned.
Notes:
  • When exporting a data set, any objects that are not used in the data set are not contained in the exported data set. Therefore, they are not included when the data set is imported.

    For example, if the object or label "car" is defined but is not used in any of the images in the data set, the exported data set does not include the "car" object or label. When the data set is imported, the "car" object or label is not created.

  • In PowerAI Vision 1.1.1, any information about augmented images is lost on export. Therefore, if the data set is later imported (regardless of the product version), the augmented images will be in the data set, but they will no longer be marked as augmented.
Export a model
When you export a model, a zip file is generated that contains the model and some additional files, depending on the model type. The generated zip file is not encrypted or password protected.

To export a model, open the Models page, select the model you want to export, then click Export in the left pane. The model is saved in your default download directory as character_string.zip; where character_string is randomly generated by the system.

Note:

If the model is not a Custom model that was imported from the Models page, the exported model can only be used with PowerAI Vision. It can be imported into the Inference Server product and deployed with the Inference Server product.

It is not recommended that you use an exported model with an earlier version of the product than it was exported from. Additionally, a model from a prior version will not have support for features that were added to later versions of the product. That is, if you export a model from version x.1 and import it into x.2, features that were added in x.2 will not be supported on the imported model.

Export the results of action detection or object detection inference
You can download the results from an inference that was run in the last seven days on an action detection or object detection model by following these steps:
  1. Open the Deployed model page, then click the name of the model.
  2. Scroll to the Test Model section and click Results history. A window opens that shows all inference results from the past seven days.
  3. Select the results that you want to view and click Load results.
  4. Scroll down to any results that you want to download and click Export result.
    • For action detection: A ZIP file is downloaded to your system. This file contains the original video, a CSV file that contains the result information, and if the option to generate the annotated video was selected when the inference operation was started, the processed video with action labels added as annotations.
    • For object detection: A ZIP file is downloaded to your system. This file contains the original video, a JSON file that contains the result information, and the processed video with object labels added as annotations.

Downloading assets

When you train certain types of models, additional downloadable assets are generated. To download these assets, open the Deployed models page and click the appropriate model.
Core ML assets
When you download Core ML assets, the model is downloaded as an .mlmodel file. This file can be deployed onto an Xcode project and can be used for inference directly on the device.
TensorRT assets
When you download TensorRT assets, the model is downloaded as a .tar.gz file.

Importing

Import a data set
  1. Navigate to the Data sets page.
  2. Drag and drop an exported data set .zip file onto the Create box.
    Important: After the upload starts, do not close the PowerAI Vision tab or refresh the page. Doing so stops the upload.
  3. After the upload completes, the data set has its original name.
Notes:
  • In PowerAI Vision 1.1.1, any information about augmented images is lost on export. Therefore, if the data set is later imported (regardless of the product version), the augmented images will be in the data set, but they will no longer be marked as augmented.
  • The data set associated with a model is not preserved when it is exported. Therefore, for imported models, the Data set field is set to "Not found".
Import a model

Instead of using PowerAI Vision to train a new model, you can import a model that was previously trained with PowerAI Vision, and was then exported. This lets you streamline data processing by offloading training tasks and allowing you to reuse models on multiple systems. After the model is imported to the Models page, you can deploy it in PowerAI Vision. Use the information in this topic to prepare a custom model that will be deployed in PowerAI Vision: Preparing a model that will be deployed in PowerAI Vision.

  1. Navigate to the Models page.
  2. Drag and drop a previously exported model .zip file onto the Import box.
    Important: After the upload starts, do not close the PowerAI Vision tab or refresh the page. Doing so stops the upload.
  3. After the upload completes, the model has its original name.
Parent topic: Training and working with models
Previous topic: Refining a model
Next topic: PowerAI Vision REST APIs