1. Introduction
COM6503: 3D Computer Graphics: Assignment 1 (40%)
Dr Steve Maddock
Deadline: 3pm, Wednesday 11 December
The assignment will involve using modern OpenGL to render a scene. Scene graphs are required in the modelling process and animation controls are required for hierarchical models.
2. Learningoutcomes
After completing this assignment, you will be able to:
• Use data structures and mathematics in representing and manipulating 3D objects
• Produce interactive software that makes use of a graphics API
Figure 1. The scene – a room on a spaceship with a view of space out of the window. The dashed blue arrows show the rotation of the spotlight on top of the small robot (robot 2) and the rotation of the globe.
1
Name
3. Requirements
Figure 1 shows a scene set in a room on a spaceship. The whole scene should be modelled using transformed flat planes, cubes and spheres. The view through the window is space (i.e. stars, etc).
You must satisfy all the following requirements.
3.1 The room
• In Figure 1, the room is made of three walls and a floor. The wall nearest to the viewer is not modelled. The room size is your choice.
• The left wall has a window in it.
• The back wall has a texture added to it. This
texture includes a large shiny text message of a name. You should use part of your name for this, e.g. for me, I would choose ‘Steve’. A diffuse and a specular map will be required. The specular map will make the name part appear shiny in contrast to the rest of the wall. You must create the diffuse and specular maps. These textures should be stored in files called diffuse_name.jpg and specular_name.jpg in the assets/textures folder, where your chosen name part is used in the filename so they are easy to find and check, e.g. diffuse_steve.jpg and specular_steve.jpg.
• Therightwallhasarepeatingtextureaddedtoit. Here, a picture of a bug is used (as illustrated in Figure 2, which repeats four times in one direction and three times in the other). You should make your own choice of texture. It must repeat over the wall, i.e. the wall is not a single texture image containing multiple pictures.
• The floor should be textured. You choose a texture. However, it should also feature a series of thick lines in any colour you choose. This sequence of lines is the path a robot needs to follow.
• A ceiling is not shown but should be added to make the scene look better. This should be textured too. You choose the texture.
3.2 The window and the view
• The left wall in the room is a large window.
• An outside scene of space (i.e. stars, etc) can be seen through the window. Consider how you might do this: Should the scene be a texture map pasted onto the wall to look like a fake window and a scene? Or should there be a hole in the wall for the window and a texture map is pasted onto a plane that is a certain distance outside the window (as illustrated in Figure 1)? Should the window have a frame? Or should the window be the size of the whole wall with a simple small frame around the edges? Should a box of textures be added outside the window so a texture can be seen at all angles when looking out of the window? Or should a skybox be used that is outside the whole room rather than a
separate plane with a texture map on it?
• Depending on the approach you choose, how does it look when the camera moves position in the room when looking out of the window? (Is it possible to stand in the room and always see a
view of space outside through the window?)
• The scene outside the window should change over time whilst the program is running, e.g. the
stars and planets move.
• The quality of what you produce for this part of
the scene will be part of the marking. Some alternatives indicated above are more advanced than others. You must choose what to try.
3.3 The robots
• There are two robots, both modelled as scene graph hierarchies. Both are made up of transformed spheres and/or cubes.
• Robot 1, the dancing robot, is at the back left of the room in Figure 1.
• Robot 2, the small robot, is on the track on the floor at the right side of the room in Figure 1.
• Robot 1 should feature a base, three parts that form the leg and body, two arms and a head, as illustrated in Figure 1. Each of the parts can
Figure 3. Some poses of robot 1, the dancing robot. 2
Figure 2. A bug texture in a repeating pattern.
articulate to make the robot dance, as illustrated
in Figure 3, which shows three poses.
• The head of robot 1 should be your own design. It must have a minimum of 4 pieces. Figure 4 gives a few possible designs. Again, this is a
chance to show some creativity.
Figure 4. Some possible head designs for robot 1.
• Robot 2, the small robot, has a body, two eyes and an antenna. The antenna has a spotlight attached to it which is made of two transformed spheres, one for a small bulb and one for its casing (alternatively called its holder) – these are shaded as white and grey, respectively, in Figure 1. You can choose to vary the look of the spotlight, but it must feature a bulb and a casing for the bulb. The two together indicate which direction the spotlight is pointing in and thus where the pool of light from the spotlight should appear in the scene.
• Robot 2, the small robot, continually follows the path of grey lines on the floor, with its two eyes showing the forward direction.
• When robot 2 reaches a corner in the path, it changes direction to follow the new line.
• The hierarchy and associated transformations are more important than the quality of the individual pieces in each of the robot hierarchies. Transformed spheres and cubes must be used. I want you to demonstrate that you understand transformations and a scene graph hierarchy.
• You must texture-map both robots. You must decide on which textures to use. You cannot use the same texture(s) on each robot.
• I’ll be looking for a little creativity in the animation for each robot. For example, how should robot 1 perform its dance? How should its parts move with respect to each other? Perhaps some cartoon effects could be used, e.g. head parts appearing to enlarge or move position for parts of the dance? For robot 2, when it reaches a corner, how should it turn? Should it lean a little to one side? Or spin in some way?
3.4 Spotlight
• Robot 2 has a bright white spotlight attached to the top of its antenna. The dashed lines in Figure 1 are to illustrate where this spotlight is. These dashed lines would not be seen in the real effect!!
• The spotlight continuously rotates around the top of the antenna whilst robot 2 moves. Thus, different parts of the scene will be illuminated by the spotlight as robot 2 moves. (The spotlight stops rotating when robot 2 stops moving.)
• This is an advanced requirement as you are responsible for working out how to implement a spotlight effect.
3.5 The globe
• This is made of transformed spheres (one for the globe and one for the central axis) and a transformed cube (for the pedestal).
• The globe continually rotates about the central axis.
• The texture map for the globe must look like planet Earth or a cartoon version of it. You will need to find a texture to show this or draw a cartoon version of it yourself.
• The stand for the globe also needs to be texture mapped. Try to be creative with this.
3.6 General illumination
• The scene should be illuminated with a general world light which can be positioned anywhere in the scene.
• This general world light will illuminate all parts of the scene.
• When you switch off the general light (using an interface option – see the next section), the effects of the spotlight will be much clearer.
• (You can include more than one general light if you wish.)
• You do NOT have to do shadows. Do not worry about shadow effects. (The general world light will illuminate all polygons with a normal pointing towards it and the spotlight will illuminate all objects in the direction it is pointing in which are inside its spotlight area as there are no shadow effects to show light not reaching particular points.)
3.7 User interface
• Auser-controlledcamerashouldbepositionedin the scene. Use the camera that was given in the
3
tutorial material – the mouse can be used to change the direction the camera is pointing in, and the keys can be used to move about. Do not change the key mappings from the ones in the tutorial. If you change the key mappings, it will make it difficult to mark. It doesn’t matter that the camera can move and see outside the room.
• It should be possible to turn the general light on and off (or, more creatively, dim it, i.e. reduce the intensity) from the interface.
• It should be possible to turn the spotlight (lamp bulb) on and off (or, more creatively, dim it, i.e. reduce the intensity) from the interface.
• Robot 1’s dancing animation should start when robot 2 is ‘near’ to it and stop when robot 2 is not ‘near’ to it. You decide how near it needs to be. You will need to implement a proximity test (based on distance) to do this.
• You should also add buttons on the interface, so you can start and stop robot 1’s dancing animation whenever you want to, irrespective of the proximity test. This will be useful for testing purposes rather than waiting for robot 2 to traverse all the way round the room!!
• It should be possible to stop and start robot 2’s movement - the spotlight stops rotating when robot 2 stops moving. You need to implement interface controls to do this.
• There is no need to stop and start the globe rotation. This just continually rotates.
3.8 Animation
• Some of the animations are not straightforward and you may decide not to do them, although that would affect your marks for animation.
• Use Euler angles for the animation. Do not consider using quaternions, as this is beyond the requirements for this assignment.
4. Deliverables
• You should submit a zip file containing a copy of your program code (and any other necessary resources, e.g. image files for the textures and a readme.txt file that describes everything) via Blackboard – this can be done via the link to the assignment handout.
• You should submit whatever you have done, even if you have not completed all the requirements – for example, you might have produced a model of the scene but not done the
animation. If you submit nothing, you cannot
receive any marks.
• The program MUST compile and run from the command window on a Windows PC or the terminal window on a Mac. You should assume that the jogl environment (and paths) has already been set up (on my machine), so you do not have to include this as part of what you hand in. I won’t install ‘YetAnotherIDE’ to make your program work; I want to compile and run the program from a command (or terminal) window using the standard javac and java commands.
• You must include appropriate comments to identify parts of the code that you wrote, e.g. /* I declare that this code is my own work */ /* Author <insert your name here> <insert your email address here> */. This could be done around major chunks of code and/or at the start of a class to identify the main changes you made.
• YoucanmakeuseofallthecodethatIhavegiven you in the tutorial material. However, use your comments to state which bits/chunks/files are new.
• The body of the Blackboard submission message should state that the work you have handed in is your own in addition to the code that was supplied in the tutorial material.
• The name of the main class in your program should be Spacecraft. That way it is easy for me to compile and run the program. (In previous years, where this has been ignored, I have wasted time for some handins trying to work out which was the main class to use.)
• Optional: You might like to make a short video of your animation. If you do so, DO NOT include this in the handin as it will be too big for Blackboard to handle – we tried using Blackboard for this in the past and it crashed the system!! Instead, put the animation on YouTube or your personal website and give the URL of the animation in a readme.txt file. Indeed, if you are thinking of a career in the graphics/games industry, then you should be adding such animation pieces to your personal website (your digital portfolio) to show off what you are capable of.
5. Marking
I will check that the program meets the requirements listed above. The program must compile and do some part of the work requested
4
even if it is not complete. Your program will be run and exercised thoroughly.
In considering the requirements, four aspects will be considered (including the quality of the work for each aspect):
• (29marks)Modellingthescene:Eachrobotmust be a hierarchical model. Is there a spotlight model attached to robot 2’s antenna? Is there a globe? How is the room modelled? (Consider drawing scene graphs for the scene before starting to program.)
• (28 marks) Texturing: Use of texture mapping in the scene, e.g. basic texture mapping, use of diffuse and specular textures, extra texturing effects, e.g. the changing view through the window.
• (15 marks) Lighting and interface controls: all lights should behave correctly such that their effect is seen on the scene. All necessary interface controls, as described in the above specification, should be included.
• (28 marks) Robot 1 and 2 animations. Spotlight animation. Animation for the globe. Is all the animation smooth rather than jerky?
6. Unfairmeans
• The School’s student handbooks (UG and PGT) give detailed information on the topic of unfair means and what happens if unfair means is used.
• Some students in previous years have placed solutions of their assignments on their personal world-readable websites/GitHub – where possible, they have been asked to take these down. Be careful you are not attracted to these, as using any of their code would be regarded as use of unfair means – this has happened in previous years and students have failed the module because of doing this.
7. Latesubmission
• Standard School rules will be applied if the work is handed in late – see UG and PGT handbooks.
Links to handbooks
相关推荐
Win本地部署DeepSeek
Ollama 提供了简单的接口,用于创建、运行和管理这些模型,同时还维护了一个丰富的预构建模型库,可以轻松集成到各种应用程序中。重启ollama,具体操作为打开搜索ollama进程,将其结束进程,之后再执行相关模型下载命令即可,或重启电脑!Ollama支持多种DeepSeek模型版本,用户可以根据硬件配置选择合适的模型。配置添加系统环境变量OLLAMA_MODELS,设置其值为新的路径,点击确定应用。安装完成后,在终端输入以下命令,检查Ollama版本。:14B、32B或70B版本,适合高端GPU。
COM6503: 3D Computer Graphics: 1Java
6503: 3DComputerGraphicsAssignment1(40%)11•!!•OptionalDO NOT!
全志 mipi屏幕和触摸驱动调试记录
mipi屏幕与触摸调试记录
经纬管网三维建模系统MagicPipe3D V3.8.2发布,支持PBR材质导出
经纬管网建模系统MagicPipe3D,基于二维矢量管线管点数据本地离线参数化构建地下管网三维模型(包括管道、接头、井篦、附属设施等),输出标准3DTiles、Obj模型等格式,广泛支持Cesium、Unreal、Unity、飞渡科技等引擎加载进行三维可视化、语义查询、专题分析!详见:https://blog.youkuaiyun.com/Giser_3D/article/details/154820844?"1、支持Pbr材质,可以设置粗糙和金属系数",
头歌操作系统 课后作业4.3:段页式内存管理 答案 无解析
头歌操作系统 课后作业4.3:段页式内存管理 答案 无解析
基于 ESP32 的智能药盒提醒装置设计与实现
本装置基于 ESP32 实现了智能药盒的核心功能,硬件选型兼顾成本与实用性,软件设计模块化、易拓展。后续可通过 ESP32 的 WiFi 模块接入物联网平台(如阿里云、OneNet),实现远程设置提醒、服药状态反馈等功能,进一步提升智能化水平。该装置能有效解决老人忘服药问题,具有较高的实用价值与推广潜力。
【3D仿真数据集】从“能生成”到“能用”:面向工业仿真的3D资产自治自愈流水线全链路落地
AI生成的GLB、URDF、3DGS资产普遍存在原点偏移、法线反转、贴图失效、关节超限、渲染瑕疵等天然缺陷,导致“生成一百个,能用不到一半”。通过离线入库预检 + 运行时动态自愈双层防护体系,实现多格式资产(GLB/URDF/SPZ)自动化质检、缺陷修复、全链路溯源、故障沉淀、评测基准自生长,真正解决「AI生成3D不可用、仿真易崩坏、人工治理成本高」的行业痛点。缺少一套覆盖多资产格式、事前+事后双层防护、可自治巡检、可自愈修复、可沉淀评测基准的完整3D生产流水线。
YOLO11改进-注意力-引入自调制特征聚合模块SMFA
SMFA(自调制特征聚合模块): SMFA模块用于协同建模局部和非局部信息,它分为两个分支:一个是EASA(Efficient Approximation of Self-Attention,简化的自注意力分支),用于捕获非局部信息;另一个是LDE(Local Detail Estimation,局部细节估计分支),用于捕获局部细节。EASA通过对输入特征进行下采样,然后利用全局特征的方差进行调制,再与原始特征进行聚合,提取非局部结构信息。LDE分支则通过卷积操作提取输入特征中的高频局部信息。
Cesium 建筑单体化、3DTiles 分类、楼层显隐、分户高亮(BIM + 倾斜孪生高频)
拿到 batchId,读取 BatchTable 里面的名称、面积、户主、房间用途,弹出信息面板,孪生项目标配。Revit 导出 3DTiles 常用工具:FME、CesiumLab、ifc2tiles;扩展业务:做多层勾选(多选楼层),shader 里面用。,否则 BatchTable 为空,前端无法筛选。很多人单体化失败就是字段名大小写不匹配。直接丢弃不在目标楼层的片元;
YOLOv8模型改进、剪枝、蒸馏和Web端界面的直观展示
模型蒸馏与性能提升 为了进一步优化YOLOv8模型,可以使用蒸馏技术将大型模型的知识转移到小型模型中。在模型蒸馏过程中,通过定义适当的损失函数和训练策略,学生网络可以学习教师网络的知识,并达到接近教师网络性能的效果。这样做不仅可以减少模型的存储和计算开销,还可以提高模型的推理速度和实时应用的响应能力。这样,用户可以直观地观察到模型的性能和应用的效果。为了优化YOLOv8模型并实现直观地观看检测效果或训练结果的全家通应用,本文将探讨如何通过改进、剪枝和蒸馏技术,并结合Web端界面来提升模型性能。
Unreal5从入门到精通之Landscape地形系统
虚幻引擎 5 (UE5) 能够利用其强大的地形编辑工具套件创建大规模的地形世界。使用地形工具可以创建身临其境的户外地形,这些地形经过优化,可在多种不同设备上保持可播放的帧速率。
虚拟仿真实训基地的技术实现路径:从3D采集、资源库到仿真平台的完整链路
本文提出“五段式”虚拟仿真实训链路——采集、建模、资源、平台、展示,破解职业教育实训资源持续产出难题。通过设备选型、AI建模、统一管理与过程留痕,实现教学资源标准化、可复用、可更新。强调元数据规范、轻量化处理与验收指标,避免“只采不建”“仿真脱节”等五大工程陷阱,确保基地长效运行,支撑岗位技能对接与教学改革。
SAP B1中文培訓教材_SDK
公司内部培训资料,对开发工具 ,DIAPI, UIAPI,Add-ON有详细解说
单细胞免疫组库分析3:从 scBCR-seq 序列到 3D 结构-IgFold 与 PyMOL 全流程深度解析
本文手把手教程基于scBCR-seq数据,利用IgFold预测抗体Fv区三维结构,并结合PyMOL完成从序列着色、CDR注释到氢键网络可视化与SCI级高颜值出图的全流程。重点解决Windows下IgFold安装难题,通过Annotate_v插件精准标注互补决定区,借助界面分析与半透明表面渲染,实现抗体-抗原结合位点的精细解析,助力免疫学研究与高水平论文发表。
计算机控制技术及应用
《计算机控制技术及应用》通过总结作者的科研成果和吸收国内外的先进理论、方法和技术,以计算机控制系统的网络化、开放化、智能化和集成化发展趋势为主线索,主要介绍计算机控制系统的基本概念和计算机控制系统的典型形式,计算机控制系统的数学描述与基本分析方法,以及数字PID控制器、最少拍控制系统、达林算法、状态变量设计法等设计方法及其各种改进算法。重点突出计算机控制系统输入输出通道接口的设计方法和典型控制系统的设计案例与实现方法等。
最新的sdio 4.0 spec 协议标准 带书签 高清完整PDF版
Part-E1-SDIO-Specification-Ver4.00-Final-120220.pdf:最新的sdio 4.0 spec 协议标准 带书签 高清完整PDF版,包括SDIO 接口的外设,SDIO 接口官方协议等。
AI生成的3D场景只想改一处,用重新生成、自然语言编辑还是DCC手修?按5类改动选
AI 生成的 3D 场景出现问题后,该重新生成、用自然语言编辑,还是进入 Blender 手修?本文从错误范围、精度要求、依赖关系、回退能力和目标端验收五个角度判断,针对整体体量、局部造型、重复部件、拓扑与 UV、绑定与制造结构给出选择矩阵,并用三轮小样验证方法,帮助创作者控制返工成本。
文档:亲测可用解决vnc灰屏,黑屏,鼠标大黑叉情况
主要解决vnc-viewer灰屏,黑屏,鼠标大黑叉等现象,详细见文档。
陕西园区 IBMS集成 3D 数字孪生平台DDC楼宇自控:和 IBMS 有什么区别
交通枢纽是最考验“空间感”的建筑:航站楼、换乘通道、行李系统、机房、管廊,动辄几十万平方米,设备藏在吊顶里、竖井里、夹层里。映射关系是核心工作:每一个点位要绑定到模型里的对应设备或空间,比如“三层东区 AHU-3-2 送风温度”要落到那台机组的三维位置上。培训价值:新员工在模型里熟悉设备位置与系统关系,比翻图纸、跑现场快得多,尤其适合人员流动大的物业。模型与台账同步:设备更换后模型与档案一起更新,否则三维画面会逐渐“失真”,一年之后没人敢信它。可分期建模:先做重点区域与关键系统,再逐步补全,不必一次做完。
3D建筑导出后总绕错位置旋转?用5步检查原点、放置锚点与交付范围
3D 建筑导出后绕错位置旋转、放置偏移或缩放离地,未必只是原点设置错误。本文区分对象原点、操作枢轴与放置锚点,通过冻结交付范围、检查锚点、追踪父级变换、执行三类操作测试,以及源软件、空场景回读和目标端核对,定位首次出现差异的环节,并提供最小交付记录表,避免用拖动补偿掩盖坐标问题。
AI建筑白模交给客户、设计师和开发者,网页链接、GLB还是DCC源文件怎么选?
AI 建筑白模交给客户、设计师和开发者,不宜只发同一种交付物。本文按接收任务区分网页评审副本、GLB 等轻量交换文件和 DCC 源工程,分别说明版本定位、单位轴向、对象映射、材质依赖与编辑上下文的验收方法,并介绍如何从同一审核母版派生成果,避免多版本失配。附交付对照表与检查清单,明确 “能查看”“能集成”“能继续编辑” 的区别,以及白模不能替代正式工程资料的边界。
百炼 CLI 生成物流主题 3D 视觉概念图教程:一个关键词出主视角/鸟瞰/特写/风格变体一套图,含默认开关与水印坑实测
做物流投标配图、方案配图的人经常遇到同一个需求:手里只有一个关键词,交付物却是一整套 3D 视觉概念图。本文以「无人仓储中心」为例,用百炼命令行的两条命令搭一条两段式生图链路,一个关键词产出主视角、鸟瞰、设备特写与风格变体成套图,并把四条可选路线的等待时间、画面文字注入情况和成套一致性一次测完,给出可直接抄的接法与参数。
机器视觉 3D 相机与机器人引导全解析:四大成像原理、手眼标定与七步落地流程
3D 视觉的精度上限,80% 取决于。
项目选型 3D 相机推荐:光学体制匹配与视锥核算实战指南
项目选型 3D 相机推荐的第一决策变量不是分辨率或标称精度,而是。工程实测表明,抛光金属件评估 450nm 蓝光激光振镜体制,利用窄带滤波从物理层切除镜面反射过曝;深筐大件则必须按“最远工作距离下的 Z 向重复精度”反向倒推视锥覆盖,而非先选相机再看视野是否“差不多够用”。光学体制选错导致的点云缺失,后续任何算法都无法补全。
OpenViking 上线原生 3D 内容理解模型
OpenViking 作为业内首个理解三维世界的 Agent 上下文数据库产品,后续将陆续支持dxf、iges、sldprt、sldasm 等更多格式的导入与理解。现在接入,后续上线无需额外操作,即刻使用。🚀上手试用,前 50 文档免费🌟 开源 OpenViking:访问GitHub 仓库,为我们点亮一颗 Star!❓三分钟测出你是哪种 Agent 饲养员选择属于你的 OpenViking。
考古遗址怎么用AI生成3D分期复原?从证据台账、构件白模到网页时间轴
考古遗址的复原画面完整,不代表历史证据完整。本文介绍从资料授权、证据台账、现状白模到分期方案和网页时间轴的工作流程,区分现存事实、学术推断、教学简化与艺术想象,说明如何用稳定构件 ID 和版本映射避免阶段错绑,并提供三轮审核与上线清单,让 AI 辅助生成的候选可追溯、不确定性可辨识。
连锁门店方案怎么用AI生成3D预演?从需求表、空间白模到网页评审
连锁门店如何用 AI 辅助比较布局,并让品牌、运营与设计团队在线评审?本文从需求表和场地底稿出发,介绍双尺寸核对、区域与设备编号、固定观察点,以及从审核母版派生网页轻量版的方法,并通过品牌运营、空间资料、网页权限三轮审核保持版本对应。AI 用于生成候选与辅助表达,网页用于沟通方案,两者都不能替代现场测量、专业设计及工程审核。
把品牌官网做成一个 3D 世界:scroll-world
你向下滚动鼠标,画面里的镜头开始往前飞——先掠过一栋小楼的外立面,然后穿墙进入室内,扫过办公区,再从另一扇窗穿出去,平滑地落到下一个场景。整个过程没有黑屏、没有 loading、没有剪辑点,像一镜到底的电影镜头,而控制这个镜头的,就是你手里的滚轮。scroll-world 的画面是 AI 预先生成好的视频,滚动操作的是视频的播放进度(时间线),而不是实时驱动一个 3D 相机。这是效果上限最高的方式,但也是最贵的。如果你是对外交付的服务商,对客户报的是解决方案的价格,不是 API 的价格。
微缩场景怎么用AI生成3D?单图、分件参考还是布局草图,按4层输入减少返工
一张港口概念图能说明场景气氛,却未必交代道路、高差、建筑背面和对象边界。本文按整体构图、空间布局、对象分件、尺寸与交付四层整理 AI 生成 3D 场景的输入,比较主图、布局草图和分件参考各自的作用与局限,并提供三轮试跑方法和可复制输入卡。生成后通过检查背面、量测尺寸、单独移动房屋及重新导入目标软件,判断候选是否满足后续编辑要求。
内网部署3D虚拟展厅,渲染与等保该先解哪一环?
将浏览器端三维展厅内网部署,核心挑战非“能否打开”,而在于拓扑划分、首屏优化、权限模型、等保合规四关。建议三区隔离:接入区(反向代理)、应用区(引擎服务)、数据区(数据库与缓存)。无CDN时,以本地静态节点+多档纹理+瓦片预热替代,首屏目标依实测定。权限按三员分立设计,数据分三类存储并加密脱敏。等保2.0要求逐项配置,日志留存≥6个月,信创需锁定型号并实测国产显卡性能。验收聚焦首屏交互时间、帧率与显存占用,容量按并发峰值规划,成本含硬件、测评与运维。关键在早将合规转化为可验证配置,避免2026年项目翻车于第
2395



被折叠的 条评论
为什么被折叠?



