A collection of machine learning projects in computer vision and natural language processing.
- image - a barebones convolutional neural network trained on CIFAR-10 for image classification
- emotion - a fine-tuned SqueezeBERT model trained on Google's GoEmotions dataset and a HuggingFace emotion dataset for text emotion analysis
- translate - a basic transformer trained on a translation dataset for sequence to sequence translation
- yolo - a YOLO (You Only Look Once) model pretrained on ImageNet and trained on PASCAL's VOC dataset for object detection
Each project contains core preprocessing, training, and inference scripts, which are described below.
params.py defines high-level project constants and model hyperparameters.
dataset.py defines a custom PyTorch Dataset class definition for the dataset used and includes relevant data preprocessing methods.
model.py defines a custom PyTorch Module class definition for the model architecture used.
train.py contains the main training loop logic.
predict.py contains the prediction script for model inference.