SQL Database Tables and Stored Procedures
Below are the data tables that you will find in your database after deployment.
| Table | Description |
|---|---|
| dbo.ImageStore | FileTable, used to save the images |
| dbo.features | SQL table which used to save images' path, label and DNN features |
| dbo.training_images | training images |
| dbo.testing_images | testing images |
| dbo.evaluation_images | image pairs used to evaluate the model |
| dbo.scores | predicted scores of all the images |
| dbo.model | trained model |
| dbo.query_images | query images |
| dbo.ranking_results | ranked candidates for all the query images |
The following stored procedures are used in this solutions:
| Stored Procedure | Description |
|---|---|
| EvaluateModel | Evaluates the performance of the model in terms of ranking |
| FeaturizeImages | Generates features from the images using a pre-trained Resnet in `microsoftml` |
| Initial_Run_Once_Py | Runs the training workflow natively in SQL for this solution |
| PrepareData | Creates and prepares the training/testing/evaluation image set |
| RankCandidates | Finds similar images for each query image |
| TrainClassifier | Trains a neural network model using `microsoftml` library and saves the model into SQL table |
