PA 3: Image Captioning

Contributors

• Edwin Tse ptse@ucsd.edu
• Moses Wong mmwong@ucsd.edu
• Janet Lin xil130@ucsd.edu

Task

In this assignment, our main goal is to perform Image Captioning task using CNNs and LSTMs and an encoder-decoder architechture. That is, CNN (an encoder) will take the image as input and encode it into a vector of feature values. Then will be passed through a linear layer for providing the input to an LSTM (a decoder).

There are 2 tasks in total performed in this assignment, they are CNN + LSTM model and Resnet + LSTM model, where each model contains 3 individual sub tasks corresponding to different LSTM parameters. All of the results can be found in the task report.

An example of our prediction task looks like the following:

Prediction: A woman is riding a horse

How to run

In terminal, simply run the following code to start the modeling process

 python3 main.py

Usage

• Make sure the following config files are properly stored in the same directory: -- task-1-600emsize-config.json -- task-1-default-config.json -- task-1-1024hidden-config.json
• Simply run python3 main.py to start the experiment
• The logs, stats, plots and saved models, accuracies for each model will be generated after main.py is finished running.

Files

• main.py: Main driver class
• experiment.py: Main experiment class. Initialized based on config - takes care of training, saving stats and plots, logging and resuming experiments.
• dataset_factory.py: Factory to build datasets based on config
• model_factory.py: Factory to build models based on config
• file_utils.py: utility functions for handling files
• caption_utils.py: utility functions to generate bleu scores
• vocab.py: A simple Vocabulary wrapper
• coco_dataset.py: A simple implementation of torch.utils.data.Dataset the Coco Dataset
• get_datasets.ipynb: A helper notebook to set up the dataset in your workspace