I made this project when I joined the advanced scholarship program from IDCamp 2023 in learning Machine Learning development at Dicoding Indonesia.
The dataset I used is sourced from kaggle and contains synthetically generated images of bottles scattered around a random background.
| About | Description |
|---|---|
| Dataset | Bottles Synthetic Images Click here. |
| Data All | 25000 images. |
| Categories | Soda Bottle, Wine Bottle, Water Bottle, Plastic Bottle, Beer Bottles. |
| Data Training | 20000 Images. |
| Data Testing | 5000 Images. |
The main folder contains 25000 Images divided in 5 categories each containing 5000 images with 512 X 512 RGB resolution and JPG file extension.
This project requires pandas, matplotlib, seaborn, sklearn, tensorflow libraries. So, you may need to install these packages if you want to test it.
The process I did was as follows:
- Data Understanding
- Split Training and Test data
- Augmentation & ImageDataGenerator
- Modeling: Sequential Algorithm
- Evaluation
- Model to TF-Lite Converter
I have obtained an accuracy of 0.92 which is quite good, so I can conclude that this model can work effectively by using LSTM in the model architecture. In addition, I used the Mean Absolute Error (MAE) scale as an evaluation metric.