Image Captioning using LSTMs

An implementation of the Image Captioning LSTM architectures in torch.

Requirements

• Torch7
• penlight
• nn
• nngraph
• optim
• [json] for parsing datasets
• [cutorch] for torch gpu support
• [cunn] for nn gpu support
• Java >= 8 (for Stanford CoreNLP utilities)
• Python >= 2.7

The Torch/Lua dependencies can be installed using luarocks. For example:

luarocks install nngraph

To run the program do:

./setup-packages.sh
./setup_data.sh
th image_captioning/main.lua

To run in gpu_mode do:

th image_captioning/main.lua -gpu_mode -num_epochs 10