tqxg2018 / image-caption-pytorch

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

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.

image-caption-pytorch's Issues

covering readme.txt to readme.md

The typesetting.txt is messy, maybe cover to .md is suitable.

Models

Could you make the models available on the Google drive?