Hello, I'm Mingru Huang

I am a Master’s degree student in Wuhan University of Technology. I’m interested in research in computer vision, particularly video understanding. My research includes video Q&A, video-text retrieval, and video captioning. Meanwhile, I also have some research and practice in large language models, prompt engineering, basic operator development and optimization, knowledge graphs, and Q&A systems. I hope to make a generalized multimodal video model that is affordable, secure and trustworthy for everyone.

CV Scholar Github Twitter

News

  • Nov. 2024: Invited as a reviewer for the ICME2025 conference.
  • Aug. 2024: Incorporated a project on automotive maintenance inspection using a multimodal large model.
  • Jul. 2024: Joined the SpConv operator optimization project based on MetaX MXMACA computing platform.
  • Jun. 2024: Approved for the Chinese Software Copyright “Dermatology Clinical Feature Detection and Diagnosis System”.
  • May 2024: The paper “ST-CLIP” has been accepted at ICIC 2024 conference.
  • Jan. 2024: Invited as a reviewer for the ICME2024 conference.
  • Dec. 2023: Joined the school-enterprise cooperation program of Haluo Corporation, responsible for the AI speech generation part of it.
  • Nov. 2023: Completed the Transformer Heterogeneous Bisheng C++ Arithmetic Development Project of Huawei Crowd Intelligence Program, and was responsible for the Adam Arithmetic part of the project.
  • Sept. 2023: Join a video understanding project, the main direction is dense video captioning.

Publications

Scene Knowledge Enhanced Multimodal Retrieval Model for Dense Video Captioning

Scene Knowledge Enhanced Multimodal Retrieval Model for Dense Video Captioning

Mingru Huang, Pengfei Duan, Yifang Zhang, Huimin Chen, Jiawang Peng, Shengwu Xiong
2025 Twenty-first International Conference on Intelligent Computing (ICIC 2025)

Introducing a Memory Enhanced Visual-Speech Aggregation model for dense video captioning, inspired by cognitive informatics on human memory recall. The model enhances visual representations by merging them with relevant text features retrieved from a memory bank through multimodal retrieval involving transcribed speech and visual inputs.

Project Page PDF arXiv
LDIT: Pseudo-Label Noise Adaptation via Label Diffusion Transformer

LDIT: Pseudo-Label Noise Adaptation via Label Diffusion Transformer

Jiawang Peng, Pengfei Duan, Mingru Huang, Shengwu Xiong
2025 Twenty-first International Conference on Intelligent Computing (ICIC 2025)

We reformulate label prediction as a progressive refinement process starting from an initial random guess, and propose LDiT (Label Diffusion Transformer) for pseudo-label noise adaptation. By modeling label uncertainty through a diffusion process, LDiT enables more robust learning under noisy supervision. In addition, to effectively capture the long-range dependencies in textual data, we adopt a Transformer-based latent denoising architecture with self-attention mechanisms.

Project Page PDF arXiv
ST-CLIP: Spatio-Temporal enhanced CLIP towards Dense Video Captioning

ST-CLIP: Spatio-Temporal enhanced CLIP towards Dense Video Captioning

Huimin Chen, Pengfei Duan, Mingru Huang, Jingyi Guo, Shengwu Xiong
2024 Twentieth International Conference on Intelligent Computing (ICIC 2024)

Proposing a new factorized spatio-temporal self-attention paradigm to address inaccurate event descriptions caused by insufficient temporal relationship modeling between video frames and apply it to dense video captioning tasks.

Project Page PDF arXiv