• Chainer 2.0.0
• Cupy 1.0.0
• Pillow

1. Download flickr8k.zip, flickr30k.zip, or coco.zip from: http://cs.stanford.edu/people/karpathy/deepimagesent/
2. Extract downloaded zip file, and you'll get dataset.json and vgg_feats.mat.

$ python src/convert_dataset.py dataset.json dataset.pkl

Parameters:

• sentence JSON file of dataset.
• output pkl file.

$ python src/train.py -g 0 -s dataset.pkl -i vgg_feats.mat -o model/caption_gen

Options:

• -g, --gpu: (optional) GPU device index (default: -1).
• -s, --sentence: (required) sentence dataset file path.
• -i, --image: (required) image feature file path.
• -m, --model: (optional) input model file path without extension.
• -o, --output: (required) output model file path without extension.
• --iter: (optional) the number of iterations (default: 100).

$ python src/convert_caffemodel_to_pkl.py VGG_ILSVRC_19_layers.caffemodel vgg19.pkl

Parameters:

• caffe model file path.
• output pkl file path.

$ python src/generate_caption.py -s dataset.pkl -i vgg19.pkl -m model/caption_gen_0010.model -l image/list.txt

Options:

• -s, sentence: (required) sentence dataset file path.
• -i, --image: (required) image model file path.
• -m, --model: (required) trained model file path.
• -l, --list: (required) image path list file.

image/asakusa.jpg
image/tree.jpg

MIT License