Create the m2
conda environment using the environment.yml
file:
conda env create -f environment.yml
conda activate m2
For the evaluation metrics, Please download the evaluation.zip(BaiduPan code:wuiu) and extract it to ./evaluation
. You can also refer to the M2 Transformer.
To run the code, object features and scene features for the Sydney-Captions dataset are needed.
Please download the features.zip(BaiduPan code:sdy2) and extract it to ./datasets/Sydney_Captions/features/
.
python train.py
python test.py
@ARTICLE{10298250,
author={Meng, Lingwu and Wang, Jing and Yang, Yang and Xiao, Liang},
journal={IEEE Transactions on Geoscience and Remote Sensing},
title={Prior Knowledge-Guided Transformer for Remote Sensing Image Captioning},
year={2023},
volume={61},
number={4706213},
pages={1-13},
doi={10.1109/TGRS.2023.3328181}}