Summarizing issues in the alpha stage

832201221_侯家傲 2024-12-20 23:18:35
Course for This Assignment2401_MU_SE_FZU
Team NameNoFantasy
Assignment RequirementsTeamwork—beta Spring
Objectives of This AssignmentBeta Sprint
Other ReferencesNo Reference

In the problem summary of the Alpha phase, we deeply recognize the critical role of user experience and functional improvement in ensuring the success of the FitTrack platform. Below are the detailed issues identified regarding the frontend and functional implementation:

1. The preceding segment responds to the problem

(1) Lack of Intuitiveness

The user interface design currently struggles to intuitively reflect users' fitness tracking needs, making it difficult for users to navigate and locate key features such as workout logs, health metrics, and progress tracking.

(2) Insufficient Usability

Some UI components lack user-friendly features, which may result in inconvenience or confusion during interactions, such as accessing fitness plans or logging workouts.

2. Functional Issues

(1) Communication Features

The platform currently lacks a messaging or communication feature for users to interact with fitness trainers, teammates, or community members, limiting collaboration and support within the platform.

(2) Goal-Setting and Progress Feedback

  • Inadequate Feedback Mechanism: Users may face difficulty in receiving real-time feedback on their fitness goals or progress, reducing motivation and engagement.
  • Limited Customization: The fitness plan customization options are limited, restricting users from tailoring plans to suit their specific health or fitness needs.

3. After-Workout Feedback and Support

Lack of Issue Resolution

The absence of a clear and structured issue resolution mechanism (e.g., support for incorrect fitness data logging) may hinder user trust in the platform.

The following are specific questions and answers:

Are the problems the software needs to solve clearly defined? Are typical users and typical scenarios clearly described?

Answer: Our software is designed to provide a convenient fitness platform for students on campus. Through this platform, students can easily detect their own movements and make corresponding plans. The above description has defined the goal of the software, typical users and typical usage scenarios, and a more detailed description can be found in the PPT.

Are we meeting our goals (how many features were originally planned? Is the delivery schedule on schedule? Has the planned number of users been reached?)

Answer: Our goal is basically achieved. In the original planned functions, the login, motion planning, planning and other modules have been implemented, and the search function has been upgraded accordingly.

Has the quality of the team's software engineering improved compared to the previous phase? Where is the improvement, by how much, and how can it be measured?

Answer: The quality of the team's software engineering improved, mainly in terms of work efficiency, defect rate reduction and functional reliability, and also improved in the final commissioning phase; In the Alpha stage, team members cultivated team spirit, cooperated and communicated with each other, and the work efficiency was doubled. The number of defects found during back-end development is reduced and the reliability of the system is improved.

What is the number of users and the acceptance of important features we expect? Are we any closer? What have we learned? What would we do if history repeated itself?

Answer: During the Beta phase, we will conduct user usage surveys to check the number of users and their acceptance of important features, which is in line with our expectations. Through the experience of the Alpha phase, we came to the conclusion that when it comes to UI design, we need to coordinate with the backend to avoid overly complex designs, while at the same time designing features that do not exceed our capabilities, so that we do not need to modify the UI design later. To ensure that the development process can be more efficient and accurate to meet the needs of users

Is there enough time to plan?

Answer: Our team has plenty of time in the planning phase.

How does the team resolve differences between colleagues during the planning phase?

Answer: Team members fully exchange opinions in the project QQ group, share ideas and try to understand others' perspectives. Through full communication and discussion, and finally through voting to determine the plan.

Have you completed all the planned work? Part of the reason for not finishing?

Answer: Pretty much all done. The Search, motion, and planning modules are complete

Are there clearly defined and measurable deliverables for each task?

Answer: Yes, each task has clearly defined and measurable deliverables that are discussed and reflected in the group.

Is the overall project proceeding according to plan? What surprises have occurred in the project? What risks were not estimated at the time, and why?

Answer: The project as a whole went according to plan, but something unexpected happened. After the UI design is completed, the interactive steps are directly carried out on the real-time design platform, but it is found that the interactive code cannot be exported after the completion. In addition, the UI design does not take into account the technical limitations of the backend, resulting in the need to modify the completed UI design later. The risk of not being able to export interactive code was not assessed, mainly because team members were unfamiliar with the platform and lacked relevant experience and knowledge.

...全文
176 回复 打赏 收藏 转发到动态 举报
写回复
用AI写文章
回复
切换为时间正序
请发表友善的回复…
发表回复
本系统提出了一套面向MRI全心脏分割任务的端到端深度学习解决方案,其核心创新在于将Swin Transformer作为编码器骨干网络,充分挖掘心脏MRI图像中的全局上下文信息与局部细节特征,同时引入4通道输入机制,在传统RGB三通道图像基础上额外拼接一个点提示编码通道,支持用户通过鼠标交互在图像上自由标注前景(心脏区域)与背景点,从而将先验空间信息显式地融入网络前向传播过程,实现基于稀疏标注引导的精准分割。在解码阶段,系统采用U型对称结构,通过逐步上采样与跳跃连接逐级恢复特征图分辨率,最终输出二值分割掩膜(前景/背景),有效区分心脏区域与其他组织。训练过程中,系统选用交叉熵损失函数,配合AdamW优化器与余弦退火学习率调度策略,确保模型在训练集和验证集上稳定收敛,并在每个epoch结束后自动计算全局像素准确率、平均交并比(mIoU)、平均Dice系数、平均精确率、平均召回率及平均F1分数等多项评估指标,全方位监控模型性能。系统内置了完整的训练日志保存、损失曲线与性能曲线绘制、学习率衰减可视化等功能模块,便于用户直观分析训练过程并调优超参数。在推理应用层面,系统封装了基于Tkinter框架的图形化交互界面,用户可上传任意MRI切片图像,通过鼠标左键/右键分别添加前景/背景点,点击“执行分割”按钮后即可实时生成叠加了红色半透明掩膜的分割结果图像,支持点集的增删与重置操作,交互响应灵敏,操作逻辑直观清晰。整体而言,本系统不仅实现了从多模态数据加载、交互式标注编码、基于Transformer的分割建模到图形化推理部署的全链路覆盖,更在算法层面通过点提示引导机制与Swin Transformer结构的高效结合,显著降低了对大规模标注数据的依赖,同时提升了模型对心脏边界模糊、形态变异及邻近组织干扰的鲁棒性,为心血管疾病的计算机辅助诊断、术前规划及定量分析提供了一种兼具精度与灵活性的智能
针对心脏MRI图像中边界模糊、器官形变复杂以及标注成本高昂等痛点,本研究构建了一套以Swin Transformer为编码基座、融合人机交互机制的轻量化分割系统。该方案跳出传统全自动分割的思维定式,转而采纳“模型推理+专家微调”的协同策略,在模型输入层开辟了一条额外的点提示通道,允许操作者通过鼠标标记少量前景或背景点,将这些位置信息与图像特征并行馈入网络,从而将抽象的空间先验转化为可微分的学习信号,使得分割结果能够灵活响应个体差异与局部歧义。编码端采用基于移位窗口注意力的Transformer结构,以分层递降的分辨率捕获全局感受野下的解剖结构关联,解码端则通过逐步上采样与跨层特征拼接恢复空间细节,最终输出逐像素的二分类概率图。训练数据来自心脏MRI多切片序列,每张样本不仅包含原始影像与对应金标准掩膜,还通过随机采样前景点的方式模拟真实交互场景,迫使模型学会如何从稀疏的点监督中推断完整器官轮廓。损失函数选用标准交叉熵,用以衡量预测概率与真实标签之间的分布差异,同时引入混淆矩阵模块对训练与验证阶段的像素精度、召回率、F1分数、Dice系数及平均交并比进行逐轮次追踪,所有评估曲线均自动落盘保存,便于横向对比不同超参数配置下的性能演变规律。在工具链末端,系统配套开发了一个基于Tkinter的事件驱动型图形界面,将模型推理、点标注、结果渲染与图像交互四个环节无缝串联。用户上传图像后,可通过左键与右键分别部署正负样本点,随后系统自动完成坐标缩放、通道拼接、前向传播与掩膜重采样,最终在原始影像上叠加半透明彩色蒙层,清晰勾勒出模型判定的心脏区域。整套代码逻辑紧密、模块边界清晰,既可作为医学影像分割领域的教学范例,亦可经过少量适配迁移至其他器官或模态的交互式标注任务中,具备良好的扩展潜力与实用价值。

173

社区成员

发帖
与我相关
我的任务
社区描述
2401_MU_SE_FZU
软件工程 高校
社区管理员
  • FZU_SE_TeacherL
  • 助教-吴可仪
  • 助教-孔志豪
加入社区
  • 近7日
  • 近30日
  • 至今
社区公告
暂无公告

试试用AI创作助手写篇文章吧