• State TI Thinks Resolved
  • Replies 5 replies
  • Answers 1 answer
  • Subscribers 31 subscribers
  • Views 74 views
  • Users 0 members are here
Support feedback
Options
Options
Related

MSPM0G5187: quesiton about Char recognition AI model

Frank(EP) Liu
Expert 5090 points

Part Number: MSPM0G5187

Hi,

My customer wants to change char pattern (not 0-9) from Char recognition in MSPM0G5187 reference design.

I check below. 

https://github.com/TexasInstruments/tinyml-tensorlab/tree/r1.2

image.png

for 1, customer can modify below file.

https://github.com/TexasInstruments/tinyml-tensorlab/blob/r1.2/tinyml-modelmaker/examples/MNIST_image_classification/config_image_classification_mnist.yaml

for 2, I cannot find the example for char recognition or I miss something.

image.png

can you show me where I can find it?

 

BR,

frank

 

 

  • 0
    TI__Prodigy 400 points

    Hi Frank, I see that the customer wants to modify the input parameters of the model.

    There is no specific example that exists for changing the model parameters in image classification. They can use the screenshot that you have shown in 2, and modify any of the parameters of the image classification model such as:

    input_features=(28,28), variables=1, num_classes=10, with_input_batchnorm=True

    This can be a part of the yaml file like the ones in your second screenshot. Once you have created this file, say "model_config.yaml"; you can use this config in Line 27 of  the https://github.com/TexasInstruments/tinyml-tensorlab/blob/r1.2/tinyml-modelmaker/examples/MNIST_image_classification/config_image_classification_mnist.yaml

    And then use it as model_config: /path/to/model_config.yaml

    Regards,

    Adithya

  • 0 in reply to Adithya Thonse
    TI__Expert 5090 points

    Adithya,

    i modify model_config.yaml as below.

    input_features:(28,28)
    variables:1
    num_classes:10
    with_input_batchnorm:True

    but it shows error below.

    Traceback (most recent call last):
    File "C:\Users\a0219317\Downloads\tinyml-tensorlab-r1.2 (1)\tinyml-tensorlab-r1.2\tinyml-modelmaker\tinyml_modelmaker\run_tinyml_modelmaker.py", line 156, in <module>
    main(config)
    File "C:\Users\a0219317\Downloads\tinyml-tensorlab-r1.2 (1)\tinyml-tensorlab-r1.2\tinyml-modelmaker\tinyml_modelmaker\run_tinyml_modelmaker.py", line 96, in main
    model_runner.run()
    File "C:\ti\Python310\lib\site-packages\tinyml_modelmaker\ai_modules\vision\runner.py", line 191, in run
    self.model_training.run()
    File "C:\ti\Python310\lib\site-packages\tinyml_modelmaker\ai_modules\vision\training\tinyml_tinyverse\image_classification.py", line 404, in run
    train.run(args)
    File "C:\ti\Python310\lib\site-packages\tinyml_tinyverse\references\image_classification\train.py", line 672, in run
    main(0, args)
    File "C:\ti\Python310\lib\site-packages\tinyml_tinyverse\references\image_classification\train.py", line 501, in main
    model = models.get_model(
    File "C:\ti\Python310\lib\site-packages\tinyml_tinyverse\common\models\__init__.py", line 83, in get_model
    model_config_dict.update(dict(variables=variables, num_classes=num_classes, with_input_batchnorm=with_input_batchnorm, input_features=input_features, dual_op=dual_op))
    AttributeError: 'str' object has no attribute 'update'

    Any idea?

    BR,

    frank

  • 0 in reply to Frank(EP) Liu
    TI__Prodigy 400 points

    Did you also pass the following under the 'training' section of the config_image_classification_mnist.yaml

    model_config: /path/to/model_config.yaml

  • 0 in reply to Adithya Thonse
    TI__Expert 5090 points

    Adithya,

    i only add below.

    model_config: .\examples\MNIST_image_classification\model_config.yaml
    I change the num_classes to 2 and it can pass the compilation.

    input_features: (28,28)
    variables: 1
    num_classes: 2
    with_input_batchnorm: True
    I change it back to 10 and it can pass the compilation.

    input_features: (28,28)
    variables: 1
    num_classes: 10
    with_input_batchnorm: True
    so, I think that it works now.
    BR,
    frank
  • 0 in reply to Frank(EP) Liu
    TI__Prodigy 400 points

    Hi frank, num_classes is usually inferred from the dataset itself. If this was the only parameter you wanted to change, having 10 classes in the dataset should have done it, without requiring you to change this.