Adding Models

To add a new Model:

1. Navigate to Smart Bot > Computer Vision > Models.
2. Click Add Model. The Model Configuration window is displayed.

3. Enter the unique Model Name. 
4. Choose Dataset from the drop-down list. Datasets with the status "Labelled" are displayed in the drop-down list.

5. Enter the value for Learning Rate. The value should be between 0.0 and 1.0 

The learning rate is a hyper parameter that controls the change in model with estimated error for every model weight when the model is updated. Deep learning neural networks are trained using the stochastic gradient descent optimization algorithm. 
The learning rate controls how quickly the model is adapted to the problem. Smaller learning rate requires more training epochs, whereas larger learning rate results in rapid changes and require fewer training epochs.

6. Enter the value for Number of Iterations. The value should be greater than or equal to 100. The number of epochs define the number of times the learning algorithm will work through the entire training dataset. One epoch means that each sample in the training dataset has had an opportunity to update the internal model parameters. The value of this parameter depends on the total number of training images and batch size per image. 

For example, if you have 4000 images and Images Per Batch is set to 2, the number of iterations for 10 epochs are:
one epoch = 4000 / 2 = 2000
Number of Iterations = one epoch * number of epochs = 2000 * 10 = 20000.

7. Enter the value for Train Dataset Size. The value should be between 0.5 and 0.95. You can specify the Dataset size as a percentage of the total number of images used for the training. The value 0.5 means, 50% of the total labelled images are used for training and the rest for testing.
8. Choose Base Architecture from the drop-down list. The Object Detection Architecture is used for testing. The available parameters are MaskRCNN-50, and MaskRCNN-101.
9. Choose Images Per Batch from the drop-down list. The value indicates the total number of images which are trained in one mini batch. By default, the value is set to 2, assuming that a single GPU with 16 GB GPU memory. The value is dependent on the GPU memory. You can increase the Images per Batch, if the GPU memory is high.
10. Choose Batch Size Per Image from the drop-down list. The value indicates the number of regions to be detected per image. This parameter is used to train the RPN. The available values of this parameter are 256 or 512.

11. Optionally, specify a meaningful Description. 

12. Click Start Training.

The model is added to the Models section to train the model as per the specified advanced parameters.

After the training is completed, check the accuracy for all the labelled parameters. If it is too low, then you can again train your model by changing some of the advanced parameters.