In this article, we are going to convert the TensorFlow model to tflite model and will use it in a real time Sign language detection app. we will not cover the training part in this article btw I used the TensorFlow object detection API for that.

First, we will save the inference model from the checkpoint which we created while training our model. this article is helpful for those who are starting the TensorFlow lite. I hope it will be helpful.

Source code in the end of the article

1. First we will freeze the inference graph using TensorFlow od API.

Freezing is the process to identify and save all of the required things(graph, weights, etc) in a single file that you can easily use.

Run The command in root Directory

python models/research/object_detection/exporter_main_v2.py --input_type image_tensor --pipeline_config_path /output/exported_models/training/001/pipeline.config --trained_checkpoint_dir output/exported_models/training/001/ --output_directory output/exported_models/inference_model

2.Then we will convert the model to the tflite inference graph.

python models/research/object_detection/export_tflite_graph_tf2.py --pipeline_config_path output/exported_models/inference_model/inference_modelsaved_model/pipeline.config --trained_checkpoint_dir output/exported_models/inference_model/saved_model/checkpint --output_directory output/exported_models/tflite_infernce

3.Then we will post quantize the graph and save the tflite model.

# save this file as postQuantization.py

def representative_dataset():

for _ in range(100):

data = np.random.rand(1, 320, 320, 3)

yield [data.astype(np.float32)]

import numpy as np

import tensorflow as tf

saved_model_dir = "output/exported_models/tflite_infernce/saved_model"

converter = tf.lite.TFLiteConverter.from_saved_model(saved_model_dir)

converter.allow_custom_ops = True

converter.optimizations = [tf.lite.Optimize.DEFAULT]

converter.representative_dataset = representative_dataset

converter.inference_input_type = tf.uint8 # or tf.uint8

converter.inference_output_type = tf.uint8 # or tf.uint8

tflite_quant_model = converter.convert()

with tf.io.gfile.GFile(tf_lite_model_path, 'wb') as f:

f.write(tflite_quant_model)

Source Code

4.Write metadata into the tflite model to use with the android.

''' Writing MetaData to TfLite Model

save it as MetaDataWriterFile.py'''

from tflite_support.metadata_writers import object_detector

from tflite_support.metadata_writers import writer_utils

from tflite_support import metadata

ObjectDetectorWriter = object_detector.MetadataWriter

_MODEL_PATH = <tf_lite_model_path>

_LABEL_FILE = <label_path>

_SAVE_TO_PATH = <path_to_tflite_path/tflite_with_metadata.tflite>

writer = ObjectDetectorWriter.create_for_inference(

writer_utils.load_file(_MODEL_PATH), [127.5], [127.5], [_LABEL_FILE])

writer_utils.save_file(writer.populate(), _SAVE_TO_PATH)

# Verify the populated metadata and associated files.

displayer = metadata.MetadataDisplayer.with_model_file(_SAVE_TO_PATH)

print("Metadata populated:")

print(displayer.get_metadata_json())

print("Associated file(s) populated:")

print(displayer.get_packed_associated_file_list())

Source Code

5.Clone the Tensorflow-examples repository from the TensorFlow GitHub account.

Download the Android studio and SDK file to use the android app for detection.

git clone https://github.com/tensorflow/examples

Copy your tflite_with_metadata.tflite file and rename it as detect.tflite and save it in the app/src/main/assets/detect.tflite

Change the `TF_OD_API_INPUT_SIZE` to the model in the `app\src\main\java\org\tensorflow\lite\examples\detection\DetectorActivity.java` to 320.

Create the virtual device or connect your phone and run the object detection application successfully.

If this Article helpful then Please hit a like and subscribe to the channel to encourage us to make more videos and articles.

#### In the next article, I will share how to train the TensorFlow ssdMobilenet for object detection.

```Google Drive Link with all data including android app```

https://drive.google.com/drive/folders/1P-chkrDCMS-RIgTvfdAoYFhzHBSbqftR?usp=sharing

Download the source code from here

Check the Tensorflow documentation here

Download the APK for testing from <a href="https://drive.google.com/file/d/1UlQ-7A5yzj8CnRMTyeitIdj9WpVge9hk/view?usp=sharing">Google Drive</a>

Thanks to [David Lee](https://public.roboflow.com/object-detection/american-sign-language-letters) and Roboflow for the Dataset.