An interactive web application developed with Streamlit, designed for making predictions using various machine learning models. The app dynamically generates forms and pages from JSON configuration files. ⭐ If you found this helpful, consider starring the repo!
LIBRARIES NEEDED
Tensorflow
Keras
Keras_CV
Numpy
Matplotlib
Scikit-learn
Starting with a solid baseline using a CNN, we setup a framework for comparison of different model configurations.
Combining data-augmentation with transfer-learning techniques improved performance significantly.
Our final model consists of:
Preprocessing - Lanczos5 interpolation for resizing and upscaling
Data augmentation - Rotation + Horizontal flipping
Model (transfer-learning) - EfficientNetV2S backbone for feature-extraction and fine-tuning.
Problem Description:
Multiclass classification of facial emotions from grayscale images.
MODELS IMPLEMENTED Convolutional Neural Network (for baseline model) MobileNetV2 (for transfer-learning backbone) EfficientNetV2S (for transfer-learning backbone)
LIBRARIES NEEDED Tensorflow Keras Keras_CV Numpy Matplotlib Scikit-learn
Starting with a solid baseline using a CNN, we setup a framework for comparison of different model configurations. Combining data-augmentation with transfer-learning techniques improved performance significantly. Our final model consists of: Preprocessing - Lanczos5 interpolation for resizing and upscaling Data augmentation - Rotation + Horizontal flipping Model (transfer-learning) - EfficientNetV2S backbone for feature-extraction and fine-tuning.