Skip to content

Latest commit

 

History

History
58 lines (42 loc) · 1.62 KB

README.md

File metadata and controls

58 lines (42 loc) · 1.62 KB
Raw

Image Captioning - Attend2u

Implementation of Attend to You: Personalized Image Captioning with Context Sequence Memory Networks

  • Work in progress *

Requirements

  • Python 3.5
  • TensorFlow > 1.10
  • Other requirements - check requirements.txt

Installation

  • conda create -name attend2u python=3.5 pip
  • conda activate attend2u
  • pip install -r requirements.txt

Dataset

Download Instagram dataset.
    JSON file & Images

Save the files to ${project_root}/data

Example of /data directory structure

├── data
│   ├── caption_dataset
│   ├── hashtag_dataset
│   ├── images
│   ├── json
│   ├── resnet_pool5_features

Running Code

  1. Download ResNet trained model
        Run scripts/download_pretrained_resnet_101.sh

  2. Extract images features
        Run scripts/extract_features.sh

  3. Configure the network
        Edit model parameters, training/evaluation parameters and data paths in configs.py

  4. Train
        Run train.py
            python train.py

  5. Evaluate
        Run eval.py
            python eval.py

Original code : https://github.com/cesc-park/attend2u

Coco evaluation tools for Python 3 : https://github.com/Illuminati91/pycocoevalcap