chen-yang-liu / mlat

Official pytorch implementation of paper "Remote Sensing Image Captioning Based on Multi-Layer Aggregated Transformer"

👋 About Me

I am a Ph.D. student at LEVIR Group, Beihang University (BUAA), supervised by Prof. Zhenwei Shi (史振威，国家杰青) and Prof. Zhengxia Zou (邹征夏，国家优青海外). Previously, I received my bachelor’s degree from Beihang University in June 2021. My research interests lie in remote sensing image processing, deep learning, and multimodal learning.

📎 Homepages

• Personal Pages: https://Chen-Yang-Liu.github.io
• Google Scholar: https://scholar.google.com/citations?user=SCHOLAR_ID&user=jBnA45cAAAAJ

🔥 News

• 2024.03: One paper is accepted by TGRS (JCR Q1, IF=8.2) !
• 2024.01: One paper is accepted by TGRS (JCR Q1, IF=8.2) !
• 2023.09: One paper is accepted by TGRS (JCR Q1, IF=8.2) !
• 2023.06: One paper is accepted by TGRS (JCR Q1, IF=8.2) !
• 2023.04: One paper is accepted by IGARSS 2023 (Conference) !
• 2022.10: One paper is accepted by TGRS (JCR Q1, IF=8.2) !
• 2022.02: One paper is accepted by GRSL (JCR Q1, IF=4.8) !

mlat's Issues

Pretrained model for RSICD

Dear Chen-Yang-Liu,
I'm planning to use the results of your method for a paper I'm writing.
Can you please provide the pretrained model on RSICD, so that I can run it on a custom image and obtain a caption?
Thank you for your work,

Riccardo

Question about training

First of all I would like to thank the authors for their code, it is very helpful for my work, I have encountered two problems when trying to reproduce your paper.
1: I have finished training 17 epochs, but the best-checkpoint is obtained at epoch=2, and the BLUE4 score is only about 0.202 at this time, which is not 0.34 of the BLUE4 score obtained in your article. how do I solve this problem? Can you provide us with your training logs or the Beck-checkpoint model obtained from the training?
2:During my training I noticed that my loss was decreasing at each epoch, but it seems that the final BLUE score did not improve and was always between 0.19 and 0.22 during my training? What is the reason for this situation?
By the way, I would also like to ask how long did it take you to complete the 100 epoch training? I'm using 2080Ti for training and it seems to take 4 days to complete the training, is this reasonable?
Very much looking forward to your answer, thanks again for your perfect work!