DCP-Net is a novel framework designed to enhance remote sensing semantic segmentation through distributed collaborative perception. This project builds upon the foundational concepts introduced in the "who2com" and "when2com" frameworks.
Our DCP-Net leverages and extends the baseline established by who2com and when2com, which were developed as part of the research efforts at GT-RIPL. These projects explore the dynamics of communication and collaboration among multiple agents in perception tasks, laying the groundwork for our enhancements.
While DCP-Net draws on the initial ideas of who2com and when2com, it introduces significant improvements and optimizations specifically tailored for the challenges of remote sensing:
- Self-Mutual Information Match Module: Identifies optimal collaboration opportunities more efficiently.
- Related Feature Fusion Module: Addresses the misalignment issues inherent in multi-angle remote sensing data collection.
Our modifications and extensions are designed to optimize performance in environments characterized by limited and unreliable communication channels, a common scenario in remote sensing applications.
We acknowledge and are grateful for the pioneering work done by the creators of who2com and when2com. Their open-source projects have been instrumental in shaping the initial stages of our development.
This project is released under the same license as the original projects, allowing for free use, modification, and distribution under the terms of the MIT License.
If you find DCP-Net or the foundational work of who2com and when2com useful in your research, please consider citing the original authors along with this project:
@article{wang2023dcp, title={DCP-Net: A Distributed Collaborative Perception Network for Remote Sensing Semantic Segmentation}, author={Wang, Zhechao and Cheng, Peirui and Duan, Shujing and Chen, Kaiqiang and Wang, Zhirui and Li, Xinming and Sun, Xian}, journal={arXiv preprint arXiv:2309.02230}, year={2023} }
The data supporting the findings of this study are available from the corresponding author upon reasonable request. The data that support the findings of this study are openly available in Baidu Netdisk at https://pan.baidu.com/s/1r1R7Eq73Ea00KgklnOQNVw, accessed on [date you last accessed the data]. The repository requires an access code, which is ay77.