This is the repository for the DS4440 Practical Neural Network Course Project at Northeastern University.
-
data
This directory includes the file needed to crop the original dataset: the IMDb dataset.
-
src
This directory includes the source files required to conduct the experiments.
-
baseline
This folder contains the code for the baselines, including BERT and BEiT, both pretrained and randomly initialized.
-
experiment
This folder contains the experiments for inverse modality, where BEiT processes text and BERT processes images, both pretrained and randomly initialized.
-
contrastive
This folder includes the code to extract features from images to create 768-dimensional vectors. It also contains the model used to process these vectors, both pretrained and randomly initialized.
-
-
result
This directory includes the results and Jupyter files used to plot the graphs featured in the report.
-
what else do you need
You should create a folder named “original,” which should include:
- A CSV file for the IMDb dataset.
- A folder named “Dog_Cat,” containing all 25,000 images from the Dogs and Cats dataset.