Feihong Shen (沈飞鸿)

Master student,
School of Computer Science and Engineering,
Southeast University,
Najing, Jiangsu, P.R. China
Phone: +86 188 5165 3218
Email: feihongshen@seu.edu.cn

[Resume] [GitHub] [中文主页]

About me

I am currently an incoming Master student in PALM Lab at Southeast University, where I am supervised by Prof. Hao Chen.

I obtained my B.Sc. degree in the School of Computer Science in June 2022 from Southeast University.

Research Interests: AIGC, Computer Vision, Multi-modal, Self-supervised Learning.

Major: Software Engineering, School of Computer Science and Engineering
Awards and Honors:

The First-Class Scholarship of Si-Tech(2024);

The Second-Class Academic Scholarship(2022, 2023);

The Second Prize, “Optics Valley of China∙HuaWei Cup” The 19th China Post-Graduate Mathematical Contest in Modeling(2022);
Experiences: Teaching Assistant in computer vision(For junior students, fall, 2022).

Research Direction: Implementing camera-motion-controllable image-to-video generation based on diffusion models.

Prune and Repaint: Content-Aware Image Retargeting for any Ratio. In Annual Conference on Neural Information Processing Systems, 2024. (CCF A)
Feihong Shen, Chao Li, Yifeng Geng, Yongjian Deng and Hao Chen.
We propose semantic seamcarving to tackle the semantic-absence problem of seamcarving. To address the cropping artifacts caused by pixel shifting methods, inpainting is utilized to ensure the continuity. For unavoidable salient objects' deformation, we propose image enlargement rather than image cropping.
[pdf] [code]

Modality-Balanced Contrastive Learning for RGB-D Salient Object Detection.
Hao Chen, Feihong Shen and Lichuang Zhang.
We propose a contrastive learning framework for the fusion of RGB and depth images, and apply the framework to improve RGB-D SOD tasks and extend the framework to RGB-T modalities.
[pdf] [code]

Disentangled Cross-modal Transformer for RGB-D Salient Object Detection and Beyond. IEEE Transactions on Image Processing, 2024. (CCF A)
Hao Chen, Feihong Shen, Ding Ding, Yongjian Deng and Chao Li.
We improve the modality fusion of RGB-D SOD framework, and achieve SOTA on 8 benchmarks. Our fusion stratege also works well on RGB-D semantic segmentation tasks.
[pdf] [code]