This project involves training a Convolutional Neural Network (CNN) using TensorFlow and Keras to classify Rock-Paper-Scissors hand gestures from images.
https://github.com/dicodingacademy/assets/releases/download/release/rockpaperscissors.zip\
- Data Collection:
- Downloaded the dataset and extracted the images using Python's zipfile library.
- Data Preprocessing:
- Images were resized to 150x150 pixels to standardize the inputsize for the CNN.
- Augmented the dataset using ImageDataGenerator from TensorFlow to increase variation in the training set.
- Model Architecture:
- Designed a CNN architecture consisting of Conv2D, MaxPooling2D, Flatten, Dropout, and Dense layers.
- The final layer utilizes softmax activation for multi-class classification
- Model Training:
- Trained the model using the prepared training and validation datasets.
- Used callbacks for early stopping if validation accuracy surpassed 95% to prevent overfitting.
- Model Evaluation:
- Plotted training and validation accuracy/loss to visualize model performance.
- Achieved an accuracy of [your accuracy value]% on the validation set after [number of epochs] epochs.
- Prediction:
- Utilized the trained model to predict new Rock-Paper-Scissors images.
- Uploaded images using Google Colab and predicted the corresponding hand gesture.
Clone the repository and follow the steps provided in the Jupyter Notebook (Untitled8.ipynb) to reproduce the experiment.Use the provided trained model for predictions or further analysis.