MUC: Mixture of Uncalibrated Cameras for Robust 3D Human Body Reconstruction

Yitao Zhu*, Sheng Wang*, Mengjie Xu, Zixu Zhuang, Zhixin Wang, Kaidong Wang, Han Zhang, Qian Wang⁺

• Introduces a technique for accurately reconstructing 3D human poses and shapes from images captured by uncalibrated cameras.
• Utilizes pre-trained monocular models to estimate camera positions and employs a distance ranking optimization strategy for precise joint fusion, addressing self-occlusion issues.
• Deploys a model to reweight human surface for accurate body shape estimation.outputs.

Chatcad+: Towards a Universal and Reliable Interactive CAD using LLMs

Zihao Zhao*, Sheng Wang*, Jinchen Gu*, Yitao Zhu*, Lanzhuju Mei, Zixu Zhuang, Zhiming Cui, Qian Wang, Dinggang Shen⁺

Project |

• Integrates medical imaging and a professional knowledge base to enhance the reliability of Large Language Models in healthcare.
• Trains CLIP models on various medical imaging modalities for disease classification and designs an efficient mechanism to retrieve relevant medical expertise based on user statements.
• Uses the retrieved information to provide references, improving the trustworthiness of LLM outputs.

Melo: Low-rank Adaptation is Better than Fine-tuning for Medical Image Diagnosis

Yitao Zhu, Zhenrong Shen, Zihao Zhao, Sheng Wang, Xin Wang, Xiangyu Zhao, Dinggang Shen, Qian Wang⁺

Project |

• Transfers natural image pre-trained models to medical image diagnostic tasks using just 0.17% trainable parameters, achieving performance comparable to full model fine-tuning across various medical imaging modalities.
• Provides rapid task-switching capabilities and reduced memory usage in clinical deployment scenarios.outputs.

Doctorglm: Fine-tuning Your Chinese Doctor is not a Herculean Task

Honglin Xiong*, Sheng Wang*, Yitao Zhu*, Zihao Zhao*, Yuxiao Liu, Linlin Huang, Qian Wang, Dinggang Sheng⁺

Project |

• Developed the first Chinese medical dialogue model in China using a subset of Chinese medical dialogues, supplemented with translated high-quality English medical data and Q&A responses generated from Chinese medical textbooks.
• Employed advanced fine-tuning techniques like LoRA and p-tuning to optimize training strategies, supported by an active open-source community and enriched by over 40,000 pieces of user feedback.outputs.