page_type: reference description: ImageClassifier class for inference and exporting to tflite.

tflite_model_maker.image_classifier.ImageClassifier

<table class="tfo-notebook-buttons tfo-api nocontent" align="left"> <td> <a target="_blank" href="https://github.com/tensorflow/examples/blob/master/tensorflow_examples/lite/model_maker/core/task/image_classifier.py#L81-L344">

View source on GitHub

</a> </td> </table>

ImageClassifier class for inference and exporting to tflite.

<pre class="devsite-click-to-copy prettyprint lang-py tfo-signature-link"> <code>tflite_model_maker.image_classifier.ImageClassifier( model_spec, index_to_label, shuffle=True, hparams=hub_lib.get_default_hparams(), use_augmentation=False, representative_data=None ) </code></pre>   <table class="responsive fixed orange"> <colgroup><col width="214px"><col></colgroup> <tr><th colspan="2"><h2 class="add-link">Args</h2></th></tr> <tr> <td> `model_spec`<a id="model_spec"></a> </td> <td> Specification for the model. </td> </tr><tr> <td> `index_to_label`<a id="index_to_label"></a> </td> <td> A list that map from index to label class name. </td> </tr><tr> <td> `shuffle`<a id="shuffle"></a> </td> <td> Whether the data should be shuffled. </td> </tr><tr> <td> `hparams`<a id="hparams"></a> </td> <td> A namedtuple of hyperparameters. This function expects .dropout_rate: The fraction of the input units to drop, used in dropout layer. .do_fine_tuning: If true, the Hub module is trained together with the classification layer on top. </td> </tr><tr> <td> `use_augmentation`<a id="use_augmentation"></a> </td> <td> Use data augmentation for preprocessing. </td> </tr><tr> <td> `representative_data`<a id="representative_data"></a> </td> <td> Representative dataset for full integer quantization. Used when converting the keras model to the TFLite model with full integer quantization. </td> </tr> </table>

Methods

<h3 id="create"><code>create</code></h3>

<a target="_blank" class="external" href="https://github.com/tensorflow/examples/blob/master/tensorflow_examples/lite/model_maker/core/task/image_classifier.py#L252-L344">View source</a>

<pre class="devsite-click-to-copy prettyprint lang-py tfo-signature-link"> <code>@classmethod</code> <code>create( train_data, model_spec='efficientnet_lite0', validation_data=None, batch_size=None, epochs=None, steps_per_epoch=None, train_whole_model=None, dropout_rate=None, learning_rate=None, momentum=None, shuffle=False, use_augmentation=False, use_hub_library=True, warmup_steps=None, model_dir=None, do_train=True ) </code></pre>

Loads data and retrains the model based on data for image classification.

<table class="responsive fixed orange"> <colgroup><col width="214px"><col></colgroup> <tr><th colspan="2">Args</th></tr> <tr> <td> `train_data` </td> <td> Training data. </td> </tr><tr> <td> `model_spec` </td> <td> Specification for the model. </td> </tr><tr> <td> `validation_data` </td> <td> Validation data. If None, skips validation process. </td> </tr><tr> <td> `batch_size` </td> <td> Number of samples per training step. If `use_hub_library` is False, it represents the base learning rate when train batch size is 256 and it's linear to the batch size. </td> </tr><tr> <td> `epochs` </td> <td> Number of epochs for training. </td> </tr><tr> <td> `steps_per_epoch` </td> <td> Integer or None. Total number of steps (batches of samples) before declaring one epoch finished and starting the next epoch. If `steps_per_epoch` is None, the epoch will run until the input dataset is exhausted. </td> </tr><tr> <td> `train_whole_model` </td> <td> If true, the Hub module is trained together with the classification layer on top. Otherwise, only train the top classification layer. </td> </tr><tr> <td> `dropout_rate` </td> <td> The rate for dropout. </td> </tr><tr> <td> `learning_rate` </td> <td> Base learning rate when train batch size is 256. Linear to the batch size. </td> </tr><tr> <td> `momentum` </td> <td> a Python float forwarded to the optimizer. Only used when `use_hub_library` is True. </td> </tr><tr> <td> `shuffle` </td> <td> Whether the data should be shuffled. </td> </tr><tr> <td> `use_augmentation` </td> <td> Use data augmentation for preprocessing. </td> </tr><tr> <td> `use_hub_library` </td> <td> Use `make_image_classifier_lib` from tensorflow hub to retrain the model. </td> </tr><tr> <td> `warmup_steps` </td> <td> Number of warmup steps for warmup schedule on learning rate. If None, the default warmup_steps is used which is the total training steps in two epochs. Only used when `use_hub_library` is False. </td> </tr><tr> <td> `model_dir` </td> <td> The location of the model checkpoint files. Only used when `use_hub_library` is False. </td> </tr><tr> <td> `do_train` </td> <td> Whether to run training. </td> </tr> </table>  <table class="responsive fixed orange"> <colgroup><col width="214px"><col></colgroup> <tr><th colspan="2">Returns</th></tr> <tr class="alt"> <td colspan="2"> An instance based on ImageClassifier. </td> </tr> </table> <h3 id="create_model"><code>create_model</code></h3>

<a target="_blank" class="external" href="https://github.com/tensorflow/examples/blob/master/tensorflow_examples/lite/model_maker/core/task/image_classifier.py#L125-L138">View source</a>

<pre class="devsite-click-to-copy prettyprint lang-py tfo-signature-link"> <code>create_model( hparams=None, with_loss_and_metrics=False ) </code></pre>

Creates the classifier model for retraining.

<h3 id="create_serving_model"><code>create_serving_model</code></h3>

<a target="_blank" class="external" href="https://github.com/tensorflow/examples/blob/master/tensorflow_examples/lite/model_maker/core/task/custom_model.py#L170-L176">View source</a>

<pre class="devsite-click-to-copy prettyprint lang-py tfo-signature-link"> <code>create_serving_model() </code></pre>

Returns the underlining Keras model for serving.

<h3 id="evaluate"><code>evaluate</code></h3>

<a target="_blank" class="external" href="https://github.com/tensorflow/examples/blob/master/tensorflow_examples/lite/model_maker/core/task/classification_model.py#L53-L65">View source</a>

<pre class="devsite-click-to-copy prettyprint lang-py tfo-signature-link"> <code>evaluate( data, batch_size=32 ) </code></pre>

Evaluates the model.

<table class="responsive fixed orange"> <colgroup><col width="214px"><col></colgroup> <tr><th colspan="2">Args</th></tr> <tr> <td> `data` </td> <td> Data to be evaluated. </td> </tr><tr> <td> `batch_size` </td> <td> Number of samples per evaluation step. </td> </tr> </table>  <table class="responsive fixed orange"> <colgroup><col width="214px"><col></colgroup> <tr><th colspan="2">Returns</th></tr> <tr class="alt"> <td colspan="2"> The loss value and accuracy. </td> </tr> </table> <h3 id="evaluate_tflite"><code>evaluate_tflite</code></h3>

<a target="_blank" class="external" href="https://github.com/tensorflow/examples/blob/master/tensorflow_examples/lite/model_maker/core/task/classification_model.py#L105-L143">View source</a>

<pre class="devsite-click-to-copy prettyprint lang-py tfo-signature-link"> <code>evaluate_tflite( tflite_filepath, data, postprocess_fn=None ) </code></pre>

Evaluates the tflite model.

<table class="responsive fixed orange"> <colgroup><col width="214px"><col></colgroup> <tr><th colspan="2">Args</th></tr> <tr> <td> `tflite_filepath` </td> <td> File path to the TFLite model. </td> </tr><tr> <td> `data` </td> <td> Data to be evaluated. </td> </tr><tr> <td> `postprocess_fn` </td> <td> Postprocessing function that will be applied to the output of `lite_runner.run` before calculating the probabilities. </td> </tr> </table>  <table class="responsive fixed orange"> <colgroup><col width="214px"><col></colgroup> <tr><th colspan="2">Returns</th></tr> <tr class="alt"> <td colspan="2"> The evaluation result of TFLite model - accuracy. </td> </tr> </table> <h3 id="export"><code>export</code></h3>

<a target="_blank" class="external" href="https://github.com/tensorflow/examples/blob/master/tensorflow_examples/lite/model_maker/core/task/custom_model.py#L95-L168">View source</a>

<pre class="devsite-click-to-copy prettyprint lang-py tfo-signature-link"> <code>export( export_dir, tflite_filename='model.tflite', label_filename='labels.txt', vocab_filename='vocab.txt', saved_model_filename='saved_model', tfjs_folder_name='tfjs', export_format=None, **kwargs ) </code></pre>

Converts the retrained model based on export_format.

<table class="responsive fixed orange"> <colgroup><col width="214px"><col></colgroup> <tr><th colspan="2">Args</th></tr> <tr> <td> `export_dir` </td> <td> The directory to save exported files. </td> </tr><tr> <td> `tflite_filename` </td> <td> File name to save tflite model. The full export path is {export_dir}/{tflite_filename}. </td> </tr><tr> <td> `label_filename` </td> <td> File name to save labels. The full export path is {export_dir}/{label_filename}. </td> </tr><tr> <td> `vocab_filename` </td> <td> File name to save vocabulary. The full export path is {export_dir}/{vocab_filename}. </td> </tr><tr> <td> `saved_model_filename` </td> <td> Path to SavedModel or H5 file to save the model. The full export path is {export_dir}/{saved_model_filename}/{saved_model.pb|assets|variables}. </td> </tr><tr> <td> `tfjs_folder_name` </td> <td> Folder name to save tfjs model. The full export path is {export_dir}/{tfjs_folder_name}. </td> </tr><tr> <td> `export_format` </td> <td> List of export format that could be saved_model, tflite, label, vocab. </td> </tr><tr> <td> `**kwargs` </td> <td> Other parameters like `quantized_config` for TFLITE model. </td> </tr> </table> <h3 id="predict_top_k"><code>predict_top_k</code></h3>

<a target="_blank" class="external" href="https://github.com/tensorflow/examples/blob/master/tensorflow_examples/lite/model_maker/core/task/classification_model.py#L67-L95">View source</a>

<pre class="devsite-click-to-copy prettyprint lang-py tfo-signature-link"> <code>predict_top_k( data, k=1, batch_size=32 ) </code></pre>

Predicts the top-k predictions.

<table class="responsive fixed orange"> <colgroup><col width="214px"><col></colgroup> <tr><th colspan="2">Args</th></tr> <tr> <td> `data` </td> <td> Data to be evaluated. Either an instance of DataLoader or just raw data entries such TF tensor or numpy array. </td> </tr><tr> <td> `k` </td> <td> Number of top results to be predicted. </td> </tr><tr> <td> `batch_size` </td> <td> Number of samples per evaluation step. </td> </tr> </table>  <table class="responsive fixed orange"> <colgroup><col width="214px"><col></colgroup> <tr><th colspan="2">Returns</th></tr> <tr class="alt"> <td colspan="2"> top k results. Each one is (label, probability). </td> </tr> </table> <h3 id="summary"><code>summary</code></h3>

<a target="_blank" class="external" href="https://github.com/tensorflow/examples/blob/master/tensorflow_examples/lite/model_maker/core/task/custom_model.py#L65-L66">View source</a>

<pre class="devsite-click-to-copy prettyprint lang-py tfo-signature-link"> <code>summary() </code></pre> <h3 id="train"><code>train</code></h3>

<a target="_blank" class="external" href="https://github.com/tensorflow/examples/blob/master/tensorflow_examples/lite/model_maker/core/task/image_classifier.py#L140-L196">View source</a>

<pre class="devsite-click-to-copy prettyprint lang-py tfo-signature-link"> <code>train( train_data, validation_data=None, hparams=None, steps_per_epoch=None ) </code></pre>

Feeds the training data for training.

<table class="responsive fixed orange"> <colgroup><col width="214px"><col></colgroup> <tr><th colspan="2">Args</th></tr> <tr> <td> `train_data` </td> <td> Training data. </td> </tr><tr> <td> `validation_data` </td> <td> Validation data. If None, skips validation process. </td> </tr><tr> <td> `hparams` </td> <td> An instance of hub_lib.HParams or train_image_classifier_lib.HParams. Anamedtuple of hyperparameters. </td> </tr><tr> <td> `steps_per_epoch` </td> <td> Integer or None. Total number of steps (batches of samples) before declaring one epoch finished and starting the next epoch. If 'steps_per_epoch' is None, the epoch will run until the input dataset is exhausted. </td> </tr> </table>  <table class="responsive fixed orange"> <colgroup><col width="214px"><col></colgroup> <tr><th colspan="2">Returns</th></tr> <tr class="alt"> <td colspan="2"> The tf.keras.callbacks.History object returned by tf.keras.Model.fit*(). </td> </tr> </table>  <table class="responsive fixed orange"> <colgroup><col width="214px"><col></colgroup> <tr><th colspan="2"><h2 class="add-link">Class Variables</h2></th></tr> <tr> <td> ALLOWED_EXPORT_FORMAT<a id="ALLOWED_EXPORT_FORMAT"></a> </td> <td> `(<ExportFormat.TFLITE: 'TFLITE'>, <ExportFormat.LABEL: 'LABEL'>, <ExportFormat.SAVED_MODEL: 'SAVED_MODEL'>, <ExportFormat.TFJS: 'TFJS'>)` </td> </tr><tr> <td> DEFAULT_EXPORT_FORMAT<a id="DEFAULT_EXPORT_FORMAT"></a> </td> <td> `(<ExportFormat.TFLITE: 'TFLITE'>, <ExportFormat.LABEL: 'LABEL'>)` </td> </tr> </table>