This repository contains the codes for the Neural Network model for upvote/like prediction of memes using transfer learning. This was done as a part of Neural Networks and Applications course at NITK. Link to the project report.
Here, we train a neural network model (Google InceptionV4 Net) to classify a meme image whether it is dank or not. A meme is said to dank if the rating that the model predicts is greater than 0.5.
- Download the dataset from the link given in
./dataset_link.txt
- To extract the features from the meme images, run
python ./utils/feature_extraction.py
- To train the model on the extracted features, run
python train.py
- To do inference on an unseen dataset, run
python eval.py