SetObjectClassRecognizerTrainingOption
Sets the value of a training option for an object class recognizer.
You can set the following training options:
Key | Description | Default |
---|---|---|
batch_size
|
The number of images to train concurrently (1,2,4,8,16,32,64). Larger batch sizes result in faster training, but require more memory (more GPU memory if you are training with a GPU). This option applies only when you set recognizer_type to generation2 or generation3 . |
4 |
iterations
|
The number of training iterations to perform when training the neural network. A larger number of iterations can result in more accurate recognition, especially for recognizers that contain many similar classes. | 1000 |
recognizer_type
|
The type of object class recognizer to use. For information about the recognizer types, refer to the Media Server Administration Guide.
|
generation1
|
snapshot_frequency
|
The number of training iterations to perform before taking a snapshot of the recognizer. For example, if you run 1000 iterations you might take snapshots every 250 iterations, so that there are snapshots for 250, 500, 750, and 1000 iterations. Taking snapshots can help you find the optimum number of iterations. | No snapshots |
validation_proportion
|
The proportion of your training images to set aside to evaluate the performance of recognizer snapshots. Specify a number between 0 and 1. The default value uses 75% of your images for training and 25% for evaluation. Unless you set snapshot_frequency , all images are used for training and none for evaluation. |
0.25 |
NOTE: Changing a training option invalidates all training associated with a recognizer. After using this action, you must retrain Media Server by running the action BuildObjectClassRecognizer.
Type: synchronous
Parameter | Description | Required |
---|---|---|
recognizer
|
The name of the recognizer to configure. | Yes |
key
|
The name of the training option to set. | Yes |
value
|
The value for the training option. | Yes |
Example
This example sets the iterations
training option to 2500
:
/action=SetObjectClassRecognizerTrainingOption&recognizer=vehicles &key=iterations &value=2500
Response
<autnresponse> <action>SETOBJECTCLASSRECOGNIZERTRAININGOPTION</action> <response>SUCCESS</response> <responsedata></responsedata> </autnresponse>