ADC-image-captioning
Code for our paper: A Novel Actor Dual-Critic Model for Image Captioning, ICPR 2020
Ruchika Chavhan, Biplab Banerjee, Xiao Xiang Zhu, Subhasis Chaudhuri
Datasets
- Remote Sensing Image Captioning Dataset (RSICD)
- UC-Merced Captioning Dataset
Please make sure your data contains images and a json file containing captions corresponding to each image.
Methodology
Training
Step 1: Build vocabulary
python vocab_build.py --caption_path <path to json file> --vocab_path <path to save the vocab file>
Step 2: Train Model
To load a pretrained Actor and Value Network model and train
python train.py --data_path <path to images> --json_path <path to json file> --load_pretrained --actor_pretrained <path to actor model> --critic_pretrained <path to critic model>
To train from scratch
python train.py --data_path <path to images> --json_path <path to json file>
Citation
If you find this work useful, please cite our paper,
@inproceedings{chavhan2020novel,
author = {Ruchika Chavhan and Biplab Banerjee and Xiao Xiang Zhu and Subhasis Chaudhuri},
title = {A Novel Actor Dual-Critic Model for Remote Sensing Image Captioning},
booktitle = {International Conference on Pattern Recognition (ICPR)},
year = {2020}
}
References
A significant part of this code has been adapted from the code from Actor-Critic Sequence Training for Image Captioning