Deep Learning for Android Apps/

DL Model Using TF, Keras, and TF Lite

DL Model Using TF, Keras, and TF Lite

Our main assignment is to develop a DL model for a specific problem using TF and Keras. We’ve been given a dataset of different types of flowers. To load the dataset, we have to access the /usercode/flowers_dataset directory. This project is divided into the following six tasks:

  • Task 1: Process the provided dataset and divide it into training and testing sets. The dataset contains images of five different types of flowers: daisy, dandelion, rose, sunflower, and tulip. Hence, it’s a multiclass classification problem. The dataset is commonly used for educational purposes, tutorials, and demonstrations in ML and DL. To start working on the project, open the Jupyter Notebook, DL_for_Android_Project.ipynb.

  • Task 2: Design a DL model architecture using Keras, considering the input and output requirements.

  • Task 3: Train the model on the training set, and evaluate its performance using the appropriate metrics on the test set.

  • Task 4: Once we have a well-performing model, convert it to the TF Lite format with optimization techniques to reduce its size and improve efficiency while maintaining accuracy.

  • Task 5: Perform inference on new data, ensuring the predictions align with the expected results.

  • Task 6: Add a label file and/or relevant metadata to the TF Lite model. Verify that the label file has been successfully added and can be accessed during inference.

Present the results and insights gained from this project, highlighting the effectiveness of the model and the benefits of utilizing the TF Lite format for model deployment to resource-constrained mobile devices.