PA4 Image Captioning

Model built for RNN and LSTM (w/o pretrained weights) --> model can be specified when called in training (if not will default to LSTM)

Training works for both models

Evaluation works for both models

train.py -- script to train the model on training images. To see usage type python train.py --help on command line

get_data.py -- script to retrieve data (currently specific to data being on dsmlp-login.ucsd.edu server. To see usage type python get_data.py --help on command line

dependencies.txt -- Run pip install dependencies.txt from command line to make sure you have all the necessary packages installed to run to train and test the model

subset_data.py -- filters the list of caption ids to include only those that will actually be used during training (basically filters based on if the file is in the training image folder or not). Saves this list of filtered ids that can be used in data loading. To see usage type python subset_data.py --help on command line

• test function for validation split
• Integrate model to train and val loops
• Integrate resize, build_vocab into training
• Add hyperparam storing to tensorboard
• Add BLEU scoring to val and train
• Caption generation for validation set