gujiuxiang / language_cnn_iccv17

An Empirical Study of Language CNN for Image Captioning

This repository contains the code for the following paper:

• Gu, Jiuxiang, Gang Wang, and Tsuhan Chen. "An Empirical Study of Language CNN for Image Captioning." arXiv preprint arXiv:1612.07086 (2016).

@inproceedings{gu2017empirical,
  title={An empirical study of language cnn for image captioning},
  author={Gu, Jiuxiang and Wang, Gang and Cai, Jianfei and Chen, Tsuhan},
  booktitle={Proceedings of the IEEE International Conference on Computer Vision},
  pages={1222--1231},
  year={2017}
}

Installation

This code is written in Lua and requires Torch. If you're on Ubuntu, installing Torch and a few packages (using LuaRocks) in your home directory may look something like:

curl -s https://raw.githubusercontent.com/torch/ezinstall/master/install-deps | bash
git clone https://github.com/torch/distro.git ~/torch --recursive
cd ~/torch;
./install.sh      # and enter "yes" at the end to modify your bashrc
source ~/.bashrc
luarocks install nn
luarocks install nngraph
luarocks install image
luarocks install cutorch
luarocks install cunn
luarocks install torch
luarocks install nn
luarocks install optim
luarocks install lua-cjson
sudo apt-get install libprotobuf-dev protobuf-compiler
luarocks install loadcaffe
sudo apt-get install libhdf5-serial-dev libhdf5-dev
git clone https://github.com/deepmind/torch-hdf5
cd torch-hdf5
luarocks make hdf5-0-0.rockspec
luarocks install nninit

For training

1. Firstly, we need to some preprocessing. Head over to the COCO/ folder and run the IPython noetebook to download the dataset and do some preprocessing. The notebook will combine the train/val data together and create a very simple and small json file that contains a large list of image paths, and raw captions for each image, of the form:

[{ "file_path": "path/img.jpg", "captions": ["a caption", "a second caption of i"tgit ...] }, ...]

2. Once we have this, we're ready to invoke the prepro.py script, which will read all of this in and create a dataset (an hdf5 file and a json file) ready for consumption in the Lua code. For example, for MS COCO we can run the prepro file as follows:

python prepro.py --input_json coco/coco_raw.json --num_val 5000 --num_test 5000 --images_root coco/images --word_count_threshold 5 --output_json coco/cocotalk.json --output_h5 coco/cocotalk.h5
# This is telling the script to read in all the data (the images and the captions), allocate 5000 images for val/test splits respectively, and map all words that occur <= 5 times to a special UNK token.
# The resulting json and h5 files are about 30GB and contain everything we want to know about the dataset.

3. Put the two pretrained caffe models (the VGG16 prototxt configuration file and the proto binary of weights) somewhere (e.g. a model directory), and we're ready to train!

th train_cnn_rhw_coco.lua -input_h5 coco/cocotalk.h5 -input_json coco/cocotalk.json

The train script will take over, and start dumping checkpoints into the folder specified by checkpoint_path (default = current folder). You also have to point the train script to the VGGNet protos (see the options inside train.lua). If you'd like to evaluate BLEU/METEOR/CIDEr scores during training in addition to validation cross entropy loss, use -language_eval 1 option, but don't forget to download the coco-caption code into coco-caption directory.

For evaluation

In this case you want to run the evaluation script on a pretrained model checkpoint. I trained a decent one on the MS COCO dataset that you can run on your images. The pretrained checkpoint can be downloaded here: CNN+LSTM-based COCO, CNN+RHW-based COCO, CNN+RHW-based Flickr30K, CNN+LSTM-based Flickr30K. It's large because it contains the weights of a finetuned VGGNet. Now place all your images of interest into a folder, or run the testing images of MSCOCO, and run the eval script:

th eval.lua

Now visit results\*.html in your browser and you should see your predicted captions.