Gewis Lab · AI Research Team

Official Open-Source Homepage

Introduction

Welcome to the open-source homepage of Gewis Lab!

We are an academic research team focused on fundamental artificial intelligence research and intelligent system applications, affiliated with TJCU. Our team has long been dedicated to deep learning, machine learning, and their practical implementations in scientific and engineering problems.

Core Research Interests

• Robot perception and control
• Object detection (e.g., YOLO series) and image segmentation
• Embodied AI, especially Vision-Language-Action (VLA) models
• Time series modeling, analysis and prediction
• AI for Science (integrating physical mechanisms with data-driven approaches)

Technical Practice Highlights

In technical practice, we not only continuously optimize classic vision tasks (such as efficient detection systems based on YOLO) but also actively explore multimodal intelligence frontiers:

• Building and fine-tuning Vision-Language (VL) models for cross-modal understanding and reasoning
• Applying LoRA (Low-Rank Adaptation) and other parameter-efficient fine-tuning methods to reduce large model training costs and improve generalization ability
• Exploring VLA architectures that unify visual perception, language instructions, and robot action policies to promote the practical application of embodied intelligence

Adhering to the philosophy of "research-driven education, application-driven learning", we have accumulated rich experience in scientific research and engineering practice, and have won multiple provincial and ministerial-level scientific research awards. We are committed to openness, collaboration and innovation, and sincerely invite teachers, students, researchers and developers worldwide to follow, use and contribute to our projects!

中文简介

欢迎访问 Gewis Lab 官方开源主页！

我们是一支专注于人工智能基础研究与智能系统应用的学术团队，隶属于 TJCU。团队长期深耕深度学习、机器学习及其在科学与工程问题中的落地应用。

核心研究方向

• 机器人感知与控制
• 目标检测（如 YOLO 系列）与图像分割
• 具身智能（Embodied AI），特别是视觉-语言-动作（VLA）统一模型
• 时序数据建模、分析与预测
• 科学智能（AI for Science），融合物理机理与数据驱动方法

技术实践亮点

在技术实践方面，我们不仅持续优化经典计算机视觉任务（如基于 YOLO 的高效检测系统），还积极探索多模态智能前沿：

• 构建与微调视觉-语言（VL）模型，实现跨模态理解与推理
• 应用**低秩适配（LoRA）**等参数高效微调技术，降低大模型训练成本，提升泛化能力
• 探索VLA 架构，将视觉感知、语言指令与机器人动作策略统一建模，推动具身智能实用化

团队坚持**"以研促教、以用促学"**的理念，在科研与工程实践中积累了丰富经验，曾多次获得省部级及以上科研奖励。我们致力于开放、协作与创新，诚邀全球师生、研究者与开发者关注、使用并参与我们的开源项目！