Skip to content

使用keras实现show and tell。类似于https://github.com/oarriaga/neural_image_captioning ,但增加了Beam功能,可一次性给出beam_size个结果

Notifications You must be signed in to change notification settings

nanfengpo/ShowAndTell_keras

 
 

Repository files navigation

Image captioning model implemetned using Keras

Prerequisites

Python Environment

  • Anaconda 3
  • Extra Python modules
    • Keras >= 2.0.4
    • TensorFlow >= 1.1.0
    • tqdm
    • pycoco (fixed for python3)

Dataset

The script train_td.py uses MSCOCO dataset placed at ./COCO as follows.

./COCO
├── annotations
│   ├── captions_train2014.json
│   ├── captions_val2014.json
│   ├── image_info_test-dev2015.json
│   ├── image_info_test2014.json
│   ├── image_info_test2015.json
│   ├── instances_train2014.json
│   ├── instances_val2014.json
│   ├── person_keypoints_train2014.json
│   └── person_keypoints_val2014.json
└── images
    ├── test2014
    ├── test2015
    ├── train2014
    └── val2014

Other tools

  • git-lfs

Usage

# for training
python train_td.py

For caption generation demo, see test_predict.ipynb.

Reference

About

使用keras实现show and tell。类似于https://github.com/oarriaga/neural_image_captioning ,但增加了Beam功能,可一次性给出beam_size个结果

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Jupyter Notebook 99.1%
  • Python 0.9%