Image Caption Pytorch

Pytorch implementation of image caption problem.

Introduction

This is an implementation of image caption, based on two different papers. The two papers are:

1. Show and Tell: A Neural Image Caption Generator
2. Show, Attend and Tell: Neural Image Caption Generation with Visual Attention

The code is based on a-PyTorch-Tutorial-to-Image-Captioning.

Model Training and Testing

In order to run the code, a file called "dataset_coco.json" need to be download and put into the data folder. You can download the file here.

Training

1. run pip install -r requirement.txt
2. run chmod +x download.sh
3. run ./download.sh
4. run python create_input_files.py
5. run python train-traditional.py
   This is for the paper "Show and Tell: A Neural Image Caption Generator"
6. run python train-attention.py
   This is for the paper "Show, Attend and Tell: Neural Image Caption Generation with Visual Attention"

Testing

The testing code is tested under pycharm environment.

1. run caption-traditional.py
2. run caption-attention.py

Pretrained model

You can download the pretrained model here

1. The traiditional model, the password is yl2u.
2. The attentaion model, the password is lsv7.