{"@context":"https://schema.org","@type":"NewsArticle","generatedAt":"2026-10-24T06:00:48.434Z","headline":"LEGO-Anything研究：编码智能体可将单张照片转成可执行Blender 3D场景程序但几何自评接近随机","description":"LEGO-Anything让编码智能体从单张图像迭代编写可执行的Blender场景程序，并推出LEGO-Bench基准（208张图像、104个场景、443个注册资产），用仿真渲染提供精确3D真值。","url":"https://www.aioga.com/news/fjxvgq4f2pnypa9f5or45o1i3/","mainEntityOfPage":"https://www.aioga.com/news/fjxvgq4f2pnypa9f5or45o1i3/","datePublished":"2026-10-03T08:31:04.000Z","dateModified":"2026-10-03T08:31:04.000Z","inLanguage":"zh-CN","publisher":{"@type":"NewsMediaOrganization","name":"Aioga","url":"https://www.aioga.com"},"citation":["https://the-decoder.com/ai-agents-build-3d-scenes-from-photos-but-have-no-idea-if-they-got-it-right","https://aihot.news/items/fjxvgq4f2pnypa9f5or45o1i3"],"canonicalUrl":"https://www.aioga.com/news/fjxvgq4f2pnypa9f5or45o1i3/","directAnswer":{"@type":"Answer","text":"Aioga 编辑摘要：LEGO-Anything让编码智能体从单张图像迭代编写可执行的Blender场景程序，并推出LEGO-Bench基准（208张图像、104个场景、443个注册资产），用仿真渲染提供精确3D真值。 Aioga 将其归入「行业动态」方向，重点关注它对真实使用和行业竞争的影响。","url":"https://www.aioga.com/news/fjxvgq4f2pnypa9f5or45o1i3/","dateCreated":"2026-10-03T08:31:04.000Z","author":{"@type":"Organization","@id":"https://www.aioga.com/authors/aioga-editorial/#editorial-team","name":"Aioga Editorial Team","url":"https://www.aioga.com/authors/aioga-editorial/"}},"evidence":[{"@type":"CreativeWork","name":"The Decoder：AI News source article","url":"https://the-decoder.com/ai-agents-build-3d-scenes-from-photos-but-have-no-idea-if-they-got-it-right","datePublished":"2026-10-03T08:31:04.000Z","provider":{"@type":"Organization","name":"The Decoder：AI News","url":"https://the-decoder.com/ai-agents-build-3d-scenes-from-photos-but-have-no-idea-if-they-got-it-right"}},{"@type":"CreativeWork","name":"AIHot archive record","url":"https://aihot.news/items/fjxvgq4f2pnypa9f5or45o1i3","datePublished":"2026-10-03T08:31:04.000Z","provider":{"@type":"Organization","name":"AIHot","url":"https://aihot.news/items/fjxvgq4f2pnypa9f5or45o1i3"}}],"aggregationSource":"The Decoder：AI News","originalPublisher":{"name":"The Decoder：AI News","url":"https://the-decoder.com/ai-agents-build-3d-scenes-from-photos-but-have-no-idea-if-they-got-it-right"},"geoDeepAnswer":null,"article":{"id":"fjxvgq4f2pnypa9f5or45o1i3","slug":"fjxvgq4f2pnypa9f5or45o1i3","url":"https://www.aioga.com/news/fjxvgq4f2pnypa9f5or45o1i3/","title":"LEGO-Anything研究：编码智能体可将单张照片转成可执行Blender 3D场景程序但几何自评接近随机","title_en":"","summary":"LEGO-Anything让编码智能体从单张图像迭代编写可执行的Blender场景程序，并推出LEGO-Bench基准（208张图像、104个场景、443个注册资产），用仿真渲染提供精确3D真值。","source":"The Decoder：AI News","sourceUrl":"https://the-decoder.com/ai-agents-build-3d-scenes-from-photos-but-have-no-idea-if-they-got-it-right","aiHotUrl":"https://aihot.news/items/fjxvgq4f2pnypa9f5or45o1i3","publishedAt":"2026-10-03T08:31:04.000Z","category":"行业动态","score":58,"selected":false,"articleBody":["Coding agents can build an editable 3D scene from a single photo by writing and refining code step by step. The benchmark that comes with this work reveals where the process still breaks down: self-assessment and geometric accuracy.","A 3D reconstruction from a single photo is most useful when it exists as an executable program you can inspect, edit, and query. That's the premise behind LEGO-Anything, a project from researchers at the University of Maryland and AWS.","The approach is called \"Image-to-Code.\" A coding agent receives a single image and writes code for Blender, the widely used 3D software. Rather than generating the scene in one pass, the agent works iteratively: it writes code, runs it, looks at the result, and revises until the scene matches the original.","Because the output is a program, it captures objects, geometry, layout, and camera position explicitly. You can run, check, and modify the scene like any other piece of code.","To measure how well agents perform, the team introduces LEGO-Bench. It contains 208 images from 104 indoor and outdoor scenes and uses 443 registered assets.","Real photos don't provide a precise 3D ground truth to compare against, according to the researchers：https://lego-anything.com/. Simple synthetic scenes look unrealistic. So LEGO-Bench splits the difference by rendering its images from professionally built simulator scenes. The inputs look natural, while the exact geometry, depth, and object assignments stay hidden and serve as the answer key for automated scoring. Scene complexity can also be ramped up without changing lighting or camera settings.","The benchmark scores each scene on three axes: validity checks whether a usable scene artifact was delivered at all. Reconstruction measures how accurate the visible geometry is. Appearance captures how closely the look matches the original by re-rendering the submitted scene and comparing it pixel by pixel against the reference image.","All six tested GPT configurations delivered a working scene almost every time. Accuracy varied wildly, though. GPT-6 Astra, the best tested model, hit 53.4 percent on indoor scenes and 39.6 percent on outdoor scenes. Weaker configurations scored around 15 percent.","The more complex a scene, the more accuracy drops, and outdoor scenes are harder than interiors. When the researchers increased the models' reasoning budget, the GPT-6 variants improved a lot. Astra's score on an office test subset jumped from 32.3 to 61.8 percent.","To understand why, the researchers analyzed the agents' work steps. They found poor initial attempts, revisions that undid earlier progress, and unreliable self-assessment as the most common issues.","That last point is the most telling. When models had to pick which of two versions better matched the original, their geometric judgments landed near or below chance level. An agent basically can't tell whether its own scene has gotten better. The researchers conclude that refinement should rely on concrete measurements, not the agent's own judgment.","That insight led the authors to build LEGO-Plugin, an extension that needs no extra training. It anchors the starting scene in the reference image, swaps the unreliable self-judgment for concrete measurements, and shields correct progress from regressive edits. The plugin improved all six models. Weaker agents saw the biggest gains, with boosts up to 62.7 percent. The already strong top model gained only about two percentage points.","Finally, the team tested whether reconstructed scenes could serve as a basis for standard vision tasks. Because each scene is an executable program, object detection, segmentation, and depth estimation can be pulled directly from it.","Without any extra training, the scenes produced usable but unremarkable results across all three tasks. Object detection fared best, with the reconstructed scenes reaching roughly half the performance of the specialized model DINO：https://the-decoder.com/meta-makes-its-state-of-the-art-dinov3-image-analysis-model-available-for-commercial-projects/. For segmentation and depth estimation, the gap to specialized models like SAM 3：https://the-decoder.com/metas-sam-3-segmentation-model-blurs-the-boundary-between-language-and-vision/ and Depth Anything 3 was larger. The authors say executable scene programs from current coding agents show promise but aren't accurate enough. A big gap remains between a working result and a faithful reconstruction.","GPT-6 Astra's commanding lead in LEGO-Bench lines up with other observations. AI researcher Yoav Artzi sees the model as a major leap in spatial understanding and suspects it was trained on large amounts of 3D data like Blender scenes：https://the-decoder.com/gpt-6-astra-appears-to-show-a-step-change-in-spatial-reasoning-based-on-early-benchmarks/. 3D software makers are already gearing up for these kinds of agents. Unity has released official plugins for Claude Code and Codex：https://the-decoder.com/unity-launches-official-plugins-for-claude-code-and-openai-codex-to-stop-ai-agents-from-using-outdated-tutorials/. Other approaches skip code entirely and reconstruct scenes directly inside the model, like the Atlas world model from World Labs：https://the-decoder.com/world-labs-unveils-atlas-a-single-ai-model-that-generates-reconstructs-and-simulates-3d-worlds-from-just-a-few-photos/. Google Deepmind takes yet another route with GenCeption, using a video model for depth estimation and segmentation：https://the-decoder.com/google-deepmind-argues-video-generators-already-contain-the-world-models-computer-vision-has-been-missing/ that matches the performance of specialized models.","Stay in the loop on AI. Clear, useful, no fluff.","Follow The Decoder for AI news, background stories and expert analyses.","The Decoder：https://the-decoder.com/"],"articleImages":[{"sourceUrl":"https://the-decoder.com/wp-content/uploads/2026/10/lego-anything-generated-image-nano-banana-pro.jpg","alt":"Image description","afterParagraph":0,"url":"/media/articles/fjxvgq4f2pnypa9f5or45o1i3/43936512e91810a9.jpg"},{"sourceUrl":"https://the-decoder.com/wp-content/uploads/2026/10/lego-anything-01-overview.jpg","alt":"LEGO-Anything overview showing how a coding agent converts a single photo into an executable Blender program for editable 3D scenes","afterParagraph":1,"url":"/media/articles/fjxvgq4f2pnypa9f5or45o1i3/e4139d51cc6436f3.jpg"},{"sourceUrl":"https://the-decoder.com/wp-content/uploads/2026/10/lego-anything-13-evaluation-metrics.jpg","alt":"LEGO-Bench evaluation metrics for validity, geometric accuracy, and visual similarity","afterParagraph":5,"url":"/media/articles/fjxvgq4f2pnypa9f5or45o1i3/122cc078ef530b27.jpg"},{"sourceUrl":"https://the-decoder.com/wp-content/uploads/2026/10/lego-anything-04-reconstruction-examples.jpg","alt":"Reconstruction examples comparing GPT-6 Astra output to reference images and older models","afterParagraph":7,"url":"/media/articles/fjxvgq4f2pnypa9f5or45o1i3/5108d3703961dccd.jpg"},{"sourceUrl":"https://the-decoder.com/wp-content/uploads/2026/10/lego-anything-07-regression-example.jpg","alt":"Regression example where GPT-6 Astra degrades its own scene from 33.9 to 4.4 percent late in the process","afterParagraph":9,"url":"/media/articles/fjxvgq4f2pnypa9f5or45o1i3/a4b5a6dac4710906.jpg"},{"sourceUrl":"https://the-decoder.com/wp-content/uploads/2026/10/lego-anything-08-judge-heatmap.jpg","alt":"Heatmap showing model judgment accuracy on geometry near chance level","afterParagraph":10,"url":"/media/articles/fjxvgq4f2pnypa9f5or45o1i3/321ae1be4aef5bce.jpg"}],"mediaStatus":"ok","articleBodyZh":["编码代理可以通过逐步编写和优化代码，从单张照片构建可编辑的三维场景。随本研究附带的基准测试揭示了该过程仍然存在的问题：自我评估和几何精度。","从单张照片进行三维重建最有用的情况是它存在于一个可执行程序中，你可以检查、编辑和查询程序。这就是马里兰大学和 AWS 研究人员的 LEGO-Anything 项目的前提。","这种方法被称为“图像到代码”。编码代理接收一张图像，编写 Blender（广泛使用的 3D 软件）的代码。代理不是一次性生成场景，而是通过迭代工作：编写代码，运行代码，观察结果，并不断修改，直到场景与原图匹配。","由于输出是一个程序，它会明确地捕捉对象、几何、布局和相机位置。你可以像处理其他代码一样运行、检查和修改场景。","为了衡量代理的性能，团队引入了 LEGO-Bench。它包含来自 104 个室内和室外场景的 208 张图像，并使用了 443 个注册资产。","研究人员表示，真实照片无法提供可精确比较的三维真实值：https://lego-anything.com/。简单的合成场景看起来不真实。因此，LEGO-Bench 通过从专业构建的模拟器场景中渲染图像来折中处理。输入看起来很自然，而精确的几何、深度和对象分配保持隐藏，并作为自动评分的答案键。可以在不改变光照或相机设置的情况下增加场景复杂度。","该基准在三个轴上对每个场景进行评分：有效性检查是否交付了可用的场景工件；重建度量可见几何的准确性；外观通过重新渲染提交的场景并逐像素与参考图像比较，以衡量外观与原图的匹配程度。","所有六种测试的 GPT 配置几乎每次都能生成一个可用场景。然而，准确性差异很大。GPT-6 Astra 是测试过的表现最好的模型，在室内场景中达到了 53.4%，在室外场景中达到了 39.6%。较弱的配置得分约为 15%。","场景越复杂，准确性下降得越多，户外场景比室内场景更难。当研究人员增加模型的推理预算时，GPT-6变体的表现有了很大提升。Astra在办公测试子集上的得分从32.3%跃升至61.8%。","为了理解原因，研究人员分析了代理的工作步骤。他们发现最常见的问题是初次尝试效果差、修订撤销了之前的进展以及不可靠的自我评估。","最后一点最能说明问题。当模型必须选择两个版本中哪个更符合原始场景时，它们的几何判断几乎或低于随机水平。代理基本无法判断其自身场景是否有所改善。研究人员得出结论，改进应依赖于具体测量，而不是代理自身的判断。","这一见解促使作者构建了LEGO-Plugin，一个无需额外训练的扩展。它将起始场景锚定在参考图像中，用具体测量代替不可靠的自我判断，并保护正确进展不被倒退修改。该插件改善了所有六个模型。较弱的代理获益最大，提升高达62.7%。已经强大的顶级模型仅获得约两个百分点的提升。","最后，团队测试了重构场景是否可以作为标准视觉任务的基础。由于每个场景都是可执行程序，因此可以直接从中提取对象检测、分割和深度估计。","无需任何额外训练，场景在三个任务中都产生了可用但不出色的结果。对象检测表现最好，重构场景的性能约达到专用模型DINO的一半：https://the-decoder.com/meta-makes-its-state-of-the-art-dinov3-image-analysis-model-available-for-commercial-projects/。对于分割和深度估计，与SAM 3：https://the-decoder.com/metas-sam-3-segmentation-model-blurs-the-boundary-between-language-and-vision/ 和 Depth Anything 3等专用模型的差距更大。作者表示，当前编码代理生成的可执行场景程序显示出潜力，但还不够精确。从可用结果到忠实重建之间仍存在很大差距。","GPT-6 Astra 在 LEGO-Bench 的领先地位与其他观察结果一致。AI 研究员 Yoav Artzi 认为该模型在空间理解方面取得了重大飞跃，并怀疑它是通过大量 3D 数据（如 Blender 场景）训练的：https://the-decoder.com/gpt-6-astra-appears-to-show-a-step-change-in-spatial-reasoning-based-on-early-benchmarks/。3D 软件制造商已经在为这类智能体做准备。Unity 发布了针对 Claude Code 和 Codex 的官方插件：https://the-decoder.com/unity-launches-official-plugins-for-claude-code-and-openai-codex-to-stop-ai-agents-from-using-outdated-tutorials/。其他方法则完全跳过编码，直接在模型内部重建场景，例如 World Labs 的 Atlas 世界模型：https://the-decoder.com/world-labs-unveils-atlas-a-single-ai-model-that-generates-reconstructs-and-simulates-3d-worlds-from-just-a-few-photos/。Google Deepmind 则采用另一种路线，使用视频模型进行深度估计和分割的 GenCeption：https://the-decoder.com/google-deepmind-argues-video-generators-already-contain-the-world-models-computer-vision-has-been-missing/，其性能可与专用模型匹敌。","保持对 AI 的关注。清晰、有用，无废话。","关注 The Decoder 获取 AI 新闻、背景故事和专家分析。","解码器：https://the-decoder.com/"],"translationStatus":"translated","bodyOrigin":"source-page","editorial":{"summary":"Aioga 编辑摘要：LEGO-Anything让编码智能体从单张图像迭代编写可执行的Blender场景程序，并推出LEGO-Bench基准（208张图像、104个场景、443个注册资产），用仿真渲染提供精确3D真值。 Aioga 将其归入「行业动态」方向，重点关注它对真实使用和行业竞争的影响。","background":"背景分析：模型与研究类动态需要结合能力边界、开放方式、成本、可用性和真实任务表现判断，单项指标领先不等于已经形成稳定采用。","viewpoint":"Aioga 判断：这条动态更适合作为行业观察信号，当前信息足以建立线索，但不足以推导长期结论。","implications":"影响分析：对相关团队而言，短期应先核对来源、可用范围和实际成本，再判断是否值得接入或跟进。","nextStep":"后续观察：继续观察官方文档、实际可用性、价格变化、开发者反馈和竞品回应。","evidenceRefs":["title","summary","articleBody"],"confidence":"medium","status":"published","aiGenerated":false,"autoApproved":true,"generatedBy":"rule-safe-fallback","generatedAt":"2026-10-24T06:04:37.737Z","sourceHash":"e8d5b2a7f5685635","validation":{"passed":true,"mode":"rule-safe-fallback","checks":["schema","length","source-attribution","no-html"]}},"tags":["行业动态","The Decoder：AI News"],"translations":{"zh-CN":{"title":"LEGO-Anything研究：编码智能体可将单张照片转成可执行Blender 3D场景程序但几何自评接近随机","summary":"LEGO-Anything让编码智能体从单张图像迭代编写可执行的Blender场景程序，并推出LEGO-Bench基准（208张图像、104个场景、443个注册资产），用仿真渲染提供精确3D真值。","category":"行业动态","source":"The Decoder：AI News","aggregationSource":"The Decoder：AI News","pageTitle":"LEGO-Anything研究：编码智能体可将单张照片转成可执行Blender 3D场景程序但几何自评接近随机 - Aioga AI资讯","description":"LEGO-Anything让编码智能体从单张图像迭代编写可执行的Blender场景程序，并推出LEGO-Bench基准（208张图像、104个场景、443个注册资产），用仿真渲染提供精确3D真值。","url":"https://www.aioga.com/news/fjxvgq4f2pnypa9f5or45o1i3/","articleBody":["编码代理可以通过逐步编写和优化代码，从单张照片构建可编辑的三维场景。随本研究附带的基准测试揭示了该过程仍然存在的问题：自我评估和几何精度。","从单张照片进行三维重建最有用的情况是它存在于一个可执行程序中，你可以检查、编辑和查询程序。这就是马里兰大学和 AWS 研究人员的 LEGO-Anything 项目的前提。","这种方法被称为“图像到代码”。编码代理接收一张图像，编写 Blender（广泛使用的 3D 软件）的代码。代理不是一次性生成场景，而是通过迭代工作：编写代码，运行代码，观察结果，并不断修改，直到场景与原图匹配。","由于输出是一个程序，它会明确地捕捉对象、几何、布局和相机位置。你可以像处理其他代码一样运行、检查和修改场景。","为了衡量代理的性能，团队引入了 LEGO-Bench。它包含来自 104 个室内和室外场景的 208 张图像，并使用了 443 个注册资产。","研究人员表示，真实照片无法提供可精确比较的三维真实值：https://lego-anything.com/。简单的合成场景看起来不真实。因此，LEGO-Bench 通过从专业构建的模拟器场景中渲染图像来折中处理。输入看起来很自然，而精确的几何、深度和对象分配保持隐藏，并作为自动评分的答案键。可以在不改变光照或相机设置的情况下增加场景复杂度。","该基准在三个轴上对每个场景进行评分：有效性检查是否交付了可用的场景工件；重建度量可见几何的准确性；外观通过重新渲染提交的场景并逐像素与参考图像比较，以衡量外观与原图的匹配程度。","所有六种测试的 GPT 配置几乎每次都能生成一个可用场景。然而，准确性差异很大。GPT-6 Astra 是测试过的表现最好的模型，在室内场景中达到了 53.4%，在室外场景中达到了 39.6%。较弱的配置得分约为 15%。","场景越复杂，准确性下降得越多，户外场景比室内场景更难。当研究人员增加模型的推理预算时，GPT-6变体的表现有了很大提升。Astra在办公测试子集上的得分从32.3%跃升至61.8%。","为了理解原因，研究人员分析了代理的工作步骤。他们发现最常见的问题是初次尝试效果差、修订撤销了之前的进展以及不可靠的自我评估。","最后一点最能说明问题。当模型必须选择两个版本中哪个更符合原始场景时，它们的几何判断几乎或低于随机水平。代理基本无法判断其自身场景是否有所改善。研究人员得出结论，改进应依赖于具体测量，而不是代理自身的判断。","这一见解促使作者构建了LEGO-Plugin，一个无需额外训练的扩展。它将起始场景锚定在参考图像中，用具体测量代替不可靠的自我判断，并保护正确进展不被倒退修改。该插件改善了所有六个模型。较弱的代理获益最大，提升高达62.7%。已经强大的顶级模型仅获得约两个百分点的提升。","最后，团队测试了重构场景是否可以作为标准视觉任务的基础。由于每个场景都是可执行程序，因此可以直接从中提取对象检测、分割和深度估计。","无需任何额外训练，场景在三个任务中都产生了可用但不出色的结果。对象检测表现最好，重构场景的性能约达到专用模型DINO的一半：https://the-decoder.com/meta-makes-its-state-of-the-art-dinov3-image-analysis-model-available-for-commercial-projects/。对于分割和深度估计，与SAM 3：https://the-decoder.com/metas-sam-3-segmentation-model-blurs-the-boundary-between-language-and-vision/ 和 Depth Anything 3等专用模型的差距更大。作者表示，当前编码代理生成的可执行场景程序显示出潜力，但还不够精确。从可用结果到忠实重建之间仍存在很大差距。","GPT-6 Astra 在 LEGO-Bench 的领先地位与其他观察结果一致。AI 研究员 Yoav Artzi 认为该模型在空间理解方面取得了重大飞跃，并怀疑它是通过大量 3D 数据（如 Blender 场景）训练的：https://the-decoder.com/gpt-6-astra-appears-to-show-a-step-change-in-spatial-reasoning-based-on-early-benchmarks/。3D 软件制造商已经在为这类智能体做准备。Unity 发布了针对 Claude Code 和 Codex 的官方插件：https://the-decoder.com/unity-launches-official-plugins-for-claude-code-and-openai-codex-to-stop-ai-agents-from-using-outdated-tutorials/。其他方法则完全跳过编码，直接在模型内部重建场景，例如 World Labs 的 Atlas 世界模型：https://the-decoder.com/world-labs-unveils-atlas-a-single-ai-model-that-generates-reconstructs-and-simulates-3d-worlds-from-just-a-few-photos/。Google Deepmind 则采用另一种路线，使用视频模型进行深度估计和分割的 GenCeption：https://the-decoder.com/google-deepmind-argues-video-generators-already-contain-the-world-models-computer-vision-has-been-missing/，其性能可与专用模型匹敌。","保持对 AI 的关注。清晰、有用，无废话。","关注 The Decoder 获取 AI 新闻、背景故事和专家分析。","解码器：https://the-decoder.com/"]},"en":{"title":"LEGO-Anything Study: Coding Agents Can Convert a Single Photo into an Executable Blender 3D Scene Program, but Geometric Self-Evaluation Is Nearly Random","summary":"LEGO-Anything enables coding agents to iteratively write executable Blender scene programs from a single image, and introduces the LEGO-Bench benchmark (208 images, 104 scenes, 443 registered assets), which provides precise 3D ground truth through simulation rendering.","category":"Industry","source":"The Decoder：AI News","aggregationSource":"The Decoder：AI News","pageTitle":"LEGO-Anything Study: Coding Agents Can Convert a Single Photo into an Executable Blender 3D Scene Program, but Geometric Self-Evaluation Is Nearly Random - Aioga AI News","description":"LEGO-Anything enables coding agents to iteratively write executable Blender scene programs from a single image, and introduces the LEGO-Bench benchmark (208 images, 104 scenes, 443...","url":"https://www.aioga.com/en/news/fjxvgq4f2pnypa9f5or45o1i3/","contentTranslated":true,"translationStatus":"translated","translationRetryAt":"","translationError":"","sourceHash":"cb81ddf8f1934e86","translatedAt":"2026-10-03T18:41:05.615Z"},"ja":{"title":"LEGO-Anything研究：コーディングエージェントは単一の写真を実行可能なBlender 3Dシーンプログラムに変換できるが、ジオメトリの自己評価はほぼランダム","summary":"LEGO-Anythingは、コーディングエージェントが1枚の画像から実行可能なBlenderシーンプログラムを反復的に作成できるようにし、シミュレーションレンダリングによって正確な3Dグラウンドトゥルースを提供するLEGO-Benchベンチマーク（208枚の画像、104のシーン、443個の登録済みアセット）を公開しました。","category":"業界動向","source":"The Decoder：AI News","aggregationSource":"The Decoder：AI News","pageTitle":"LEGO-Anything研究：コーディングエージェントは単一の写真を実行可能なBlender 3Dシーンプログラムに変換できるが、ジオメトリの自己評価はほぼランダム - Aioga AIニュース","description":"LEGO-Anythingは、コーディングエージェントが1枚の画像から実行可能なBlenderシーンプログラムを反復的に作成できるようにし、シミュレーションレンダリングによって正確な3Dグラウンドトゥルースを提供するLEGO-Benchベンチマーク（208枚の画像、104のシーン、443個の登録済みアセット）を公開しました。","url":"https://www.aioga.com/ja/news/fjxvgq4f2pnypa9f5or45o1i3/","contentTranslated":true,"translationStatus":"translated","translationRetryAt":"","translationError":"","sourceHash":"cb81ddf8f1934e86","translatedAt":"2026-10-03T18:41:38.487Z"},"ko":{"title":"LEGO-Anything 연구: 코딩 에이전트는 단일 사진을 실행 가능한 Blender 3D 장면 프로그램으로 변환할 수 있지만, 기하학적 자기 평가는 무작위에 가깝다","summary":"LEGO-Anything은 코딩 에이전트가 단일 이미지로부터 실행 가능한 Blender 장면 프로그램을 반복적으로 작성하도록 하며, 시뮬레이션 렌더링을 통해 정밀한 3D 정답 데이터를 제공하는 LEGO-Bench 벤치마크(이미지 208장, 장면 104개, 등록된 에셋 443개)를 공개했습니다.","category":"업계 동향","source":"The Decoder：AI News","aggregationSource":"The Decoder：AI News","pageTitle":"LEGO-Anything 연구: 코딩 에이전트는 단일 사진을 실행 가능한 Blender 3D 장면 프로그램으로 변환할 수 있지만, 기하학적 자기 평가는 무작위에 가깝다 - Aioga AI 뉴스","description":"LEGO-Anything은 코딩 에이전트가 단일 이미지로부터 실행 가능한 Blender 장면 프로그램을 반복적으로 작성하도록 하며, 시뮬레이션 렌더링을 통해 정밀한 3D 정답 데이터를 제공하는 LEGO-Bench 벤치마크(이미지 208장, 장면 104개, 등록된 에셋 443개)를 공개했습니다.","url":"https://www.aioga.com/ko/news/fjxvgq4f2pnypa9f5or45o1i3/","contentTranslated":true,"translationStatus":"translated","translationRetryAt":"","translationError":"","sourceHash":"cb81ddf8f1934e86","translatedAt":"2026-10-03T18:41:32.704Z"},"es":{"title":"Investigación sobre LEGO-Anything: los agentes de codificación pueden convertir una sola fotografía en un programa ejecutable de una escena 3D de Blender, pero la autoevaluación geométrica se acerca a lo aleatorio.","summary":"LEGO-Anything permite que los agentes de codificación escriban de forma iterativa programas ejecutables de escenas de Blender a partir de una sola imagen, y presenta el punto de referencia LEGO-Bench (208 imágenes, 104 escenas, 443 activos registrados), que proporciona una verdad de referencia 3D precisa mediante renderizado de simulación.","category":"Industria","source":"The Decoder：AI News","aggregationSource":"The Decoder：AI News","pageTitle":"Investigación sobre LEGO-Anything: los agentes de codificación pueden convertir una sola fotografía en un programa ejecutable de una escena 3D de Blender, pero la autoevaluación geométrica se acerca a lo aleatorio. - Aioga Noticias de IA","description":"LEGO-Anything permite que los agentes de codificación escriban de forma iterativa programas ejecutables de escenas de Blender a partir de una sola imagen, y presenta el punto de re...","url":"https://www.aioga.com/es/news/fjxvgq4f2pnypa9f5or45o1i3/","contentTranslated":true,"translationStatus":"translated","translationRetryAt":"","translationError":"","sourceHash":"cb81ddf8f1934e86","translatedAt":"2026-10-03T18:46:18.822Z"},"fr":{"title":"Étude LEGO-Anything : les agents de codage peuvent convertir une seule photo en un programme de scène 3D Blender exécutable, mais leur auto-évaluation de la géométrie est proche du hasard.","summary":"LEGO-Anything permet aux agents de programmation d’écrire de manière itérative des programmes de scènes Blender exécutables à partir d’une seule image, et introduit le benchmark LEGO-Bench (208 images, 104 scènes, 443 actifs enregistrés), qui fournit une vérité terrain 3D précise grâce au rendu par simulation.","category":"Industrie","source":"The Decoder：AI News","aggregationSource":"The Decoder：AI News","pageTitle":"Étude LEGO-Anything : les agents de codage peuvent convertir une seule photo en un programme de scène 3D Blender exécutable, mais leur auto-évaluation de la géométrie est proche du hasard. - Aioga Actualités IA","description":"LEGO-Anything permet aux agents de programmation d’écrire de manière itérative des programmes de scènes Blender exécutables à partir d’une seule image, et introduit le benchmark LE...","url":"https://www.aioga.com/fr/news/fjxvgq4f2pnypa9f5or45o1i3/","contentTranslated":true,"translationStatus":"translated","translationRetryAt":"","translationError":"","sourceHash":"cb81ddf8f1934e86","translatedAt":"2026-10-03T18:46:03.547Z"},"de":{"title":"LEGO-Anything-Forschung: Codierungsagenten können ein einzelnes Foto in ein ausführbares Blender-3D-Szenenprogramm umwandeln, aber die geometrische Selbsteinschätzung liegt nahe am Zufall.","summary":"LEGO-Anything ermöglicht es Codierungsagenten, aus einem einzelnen Bild iterativ ausführbare Blender-Szenenprogramme zu schreiben, und führt den Benchmark LEGO-Bench ein (208 Bilder, 104 Szenen, 443 registrierte Assets), der mithilfe von Simulations-Rendering präzise 3D-Ground-Truth-Daten bereitstellt.","category":"行业动态","source":"The Decoder：AI News","aggregationSource":"The Decoder：AI News","pageTitle":"LEGO-Anything-Forschung: Codierungsagenten können ein einzelnes Foto in ein ausführbares Blender-3D-Szenenprogramm umwandeln, aber die geometrische Selbsteinschätzung liegt nahe am Zufall. - Aioga KI-News","description":"LEGO-Anything ermöglicht es Codierungsagenten, aus einem einzelnen Bild iterativ ausführbare Blender-Szenenprogramme zu schreiben, und führt den Benchmark LEGO-Bench ein (208 Bilde...","url":"https://www.aioga.com/de/news/fjxvgq4f2pnypa9f5or45o1i3/","contentTranslated":true,"translationStatus":"translated","translationRetryAt":"","translationError":"","sourceHash":"cb81ddf8f1934e86","translatedAt":"2026-10-03T18:46:18.588Z"},"pt-BR":{"title":"Estudo LEGO-Anything: um agente de codificação pode converter uma única foto em um programa executável de cena 3D do Blender, mas sua autoavaliação geométrica é próxima do aleatório","summary":"O LEGO-Anything permite que agentes de codificação desenvolvam iterativamente, a partir de uma única imagem, programas executáveis de cenas no Blender, e apresenta o benchmark LEGO-Bench (208 imagens, 104 cenas e 443 ativos registrados), que fornece dados de referência 3D precisos por meio de renderização simulada.","category":"行业动态","source":"The Decoder：AI News","aggregationSource":"The Decoder：AI News","pageTitle":"Estudo LEGO-Anything: um agente de codificação pode converter uma única foto em um programa executável de cena 3D do Blender, mas sua autoavaliação geométrica é próxima do aleatório - Aioga Notícias de IA","description":"O LEGO-Anything permite que agentes de codificação desenvolvam iterativamente, a partir de uma única imagem, programas executáveis de cenas no Blender, e apresenta o benchmark LEGO...","url":"https://www.aioga.com/pt-BR/news/fjxvgq4f2pnypa9f5or45o1i3/","contentTranslated":true,"translationStatus":"translated","translationRetryAt":"","translationError":"","sourceHash":"cb81ddf8f1934e86","translatedAt":"2026-10-03T18:50:43.596Z"},"ru":{"title":"Исследование LEGO-Anything: агент-кодировщик может преобразовать одну фотографию в исполняемую программу 3D-сцены Blender, однако самооценка геометрии близка к случайной.","summary":"LEGO-Anything позволяет кодирующим агентам итеративно создавать исполняемые программы сцен Blender по одному изображению и представляет бенчмарк LEGO-Bench (208 изображений, 104 сцены, 443 зарегистрированных актива), используя симуляционную визуализацию для получения точных 3D-эталонных данных.","category":"行业动态","source":"The Decoder：AI News","aggregationSource":"The Decoder：AI News","pageTitle":"Исследование LEGO-Anything: агент-кодировщик может преобразовать одну фотографию в исполняемую программу 3D-сцены Blender, однако самооценка геометрии близка к случайной. - Aioga Новости ИИ","description":"LEGO-Anything позволяет кодирующим агентам итеративно создавать исполняемые программы сцен Blender по одному изображению и представляет бенчмарк LEGO-Bench (208 изображений, 104 сц...","url":"https://www.aioga.com/ru/news/fjxvgq4f2pnypa9f5or45o1i3/","contentTranslated":true,"translationStatus":"translated","translationRetryAt":"","translationError":"","sourceHash":"cb81ddf8f1934e86","translatedAt":"2026-10-03T18:51:23.871Z"},"ar":{"title":"بحث LEGO-Anything: يمكن لوكيل الترميز تحويل صورة واحدة إلى برنامج مشهد ثلاثي الأبعاد في Blender قابل للتنفيذ، لكن التقييم الهندسي الذاتي قريب من العشوائي","summary":"يتيح LEGO-Anything لوكلاء البرمجة كتابة برامج قابلة للتنفيذ لمشاهد Blender بشكل تكراري انطلاقًا من صورة واحدة، ويقدّم معيار LEGO-Bench الذي يضم 208 صور و104 مشاهد و443 أصلًا مسجّلًا، مع توفير حقيقة أرضية ثلاثية الأبعاد دقيقة باستخدام التصيير بالمحاكاة.","category":"行业动态","source":"The Decoder：AI News","aggregationSource":"The Decoder：AI News","pageTitle":"بحث LEGO-Anything: يمكن لوكيل الترميز تحويل صورة واحدة إلى برنامج مشهد ثلاثي الأبعاد في Blender قابل للتنفيذ، لكن التقييم الهندسي الذاتي قريب من العشوائي - Aioga أخبار الذكاء الاصطناعي","description":"يتيح LEGO-Anything لوكلاء البرمجة كتابة برامج قابلة للتنفيذ لمشاهد Blender بشكل تكراري انطلاقًا من صورة واحدة، ويقدّم معيار LEGO-Bench الذي يضم 208 صور و104 مشاهد و443 أصلًا مسجّلً...","url":"https://www.aioga.com/ar/news/fjxvgq4f2pnypa9f5or45o1i3/","contentTranslated":true,"translationStatus":"translated","translationRetryAt":"","translationError":"","sourceHash":"cb81ddf8f1934e86","translatedAt":"2026-10-03T18:50:43.517Z"},"hi":{"title":"LEGO-Anything अनुसंधान: कोडिंग एजेंट एक ही तस्वीर से इसे निष्पादित करने योग्य Blender 3D दृश्य प्रोग्राम में बदल सकता है लेकिन ज्यामिति आत्म-मूल्यांकन लगभग यादृच्छिक है","summary":"LEGO-Anything कोडिंग एजेंट को एकल छवि से निष्पादन योग्य Blender दृश्य प्रोग्राम को क्रमिक रूप से लिखने में सक्षम बनाता है और LEGO-Bench बेंचमार्क (208 छवियाँ, 104 दृश्य, 443 पंजीकृत एसेट) प्रस्तुत करता है, जो सिमुलेशन रेंडरिंग के माध्यम से सटीक 3D ग्राउंड ट्रुथ उपलब्ध कराता है।","category":"行业动态","source":"The Decoder：AI News","aggregationSource":"The Decoder：AI News","pageTitle":"LEGO-Anything अनुसंधान: कोडिंग एजेंट एक ही तस्वीर से इसे निष्पादित करने योग्य Blender 3D दृश्य प्रोग्राम में बदल सकता है लेकिन ज्यामिति आत्म-मूल्यांकन लगभग यादृच्छिक है - Aioga AI समाचार","description":"LEGO-Anything कोडिंग एजेंट को एकल छवि से निष्पादन योग्य Blender दृश्य प्रोग्राम को क्रमिक रूप से लिखने में सक्षम बनाता है और LEGO-Bench बेंचमार्क (208 छवियाँ, 104 दृश्य, 443 पंजीकृ...","url":"https://www.aioga.com/hi/news/fjxvgq4f2pnypa9f5or45o1i3/","contentTranslated":true,"translationStatus":"translated","translationRetryAt":"","translationError":"","sourceHash":"cb81ddf8f1934e86","translatedAt":"2026-10-03T18:55:58.032Z"},"it":{"title":"Ricerca su LEGO-Anything: gli agenti di codifica possono trasformare una singola fotografia in un programma eseguibile per una scena 3D di Blender, ma l'autovalutazione della geometria è quasi casuale","summary":"LEGO-Anything consente agli agenti di codifica intelligenti di scrivere iterativamente programmi di scena Blender eseguibili a partire da una singola immagine e introduce il benchmark LEGO-Bench (208 immagini, 104 scene, 443 asset registrati), fornendo valori 3D esatti tramite rendering simulato.","category":"行业动态","source":"The Decoder：AI News","aggregationSource":"The Decoder：AI News","pageTitle":"Ricerca su LEGO-Anything: gli agenti di codifica possono trasformare una singola fotografia in un programma eseguibile per una scena 3D di Blender, ma l'autovalutazione della geometria è quasi casuale - Aioga Notizie IA","description":"LEGO-Anything consente agli agenti di codifica intelligenti di scrivere iterativamente programmi di scena Blender eseguibili a partire da una singola immagine e introduce il benchm...","url":"https://www.aioga.com/it/news/fjxvgq4f2pnypa9f5or45o1i3/","contentTranslated":true,"translationStatus":"translated","translationRetryAt":"","translationError":"","sourceHash":"cb81ddf8f1934e86","translatedAt":"2026-10-03T18:55:37.575Z"},"nl":{"title":"LEGO-Anything-onderzoek: codeeragenten kunnen een enkele foto omzetten in een uitvoerbaar Blender 3D-scèneprogramma, maar de geometrische zelfevaluatie is bijna willekeurig","summary":"LEGO-Anything stelt coderende agenten in staat om op basis van één afbeelding iteratief uitvoerbare Blender-scèneprogramma's te schrijven en introduceert de benchmark LEGO-Bench (208 afbeeldingen, 104 scènes, 443 geregistreerde assets), waarbij simulatie-rendering nauwkeurige 3D-groundtruth biedt.","category":"行业动态","source":"The Decoder：AI News","aggregationSource":"The Decoder：AI News","pageTitle":"LEGO-Anything-onderzoek: codeeragenten kunnen een enkele foto omzetten in een uitvoerbaar Blender 3D-scèneprogramma, maar de geometrische zelfevaluatie is bijna willekeurig - Aioga AI-nieuws","description":"LEGO-Anything stelt coderende agenten in staat om op basis van één afbeelding iteratief uitvoerbare Blender-scèneprogramma's te schrijven en introduceert de benchmark LEGO-Bench (2...","url":"https://www.aioga.com/nl/news/fjxvgq4f2pnypa9f5or45o1i3/","contentTranslated":true,"translationStatus":"translated","translationRetryAt":"","translationError":"","sourceHash":"cb81ddf8f1934e86","translatedAt":"2026-10-03T18:55:35.438Z"},"tr":{"title":"LEGO-Anything araştırması: Kodlama ajanları tek bir fotoğrafı çalıştırılabilir bir Blender 3D sahne programına dönüştürebiliyor, ancak geometriye ilişkin öz değerlendirmeleri rastgeleye yakın.","summary":"LEGO-Anything, kodlama ajanlarının tek bir görüntüden yola çıkarak yürütülebilir Blender sahne programlarını yinelemeli biçimde yazmasını sağlar ve simülasyonla oluşturulan görüntüler aracılığıyla hassas 3D gerçeği sunan LEGO-Bench kıyaslama veri setini (208 görüntü, 104 sahne, 443 kayıtlı varlık) kullanıma sunar.","category":"行业动态","source":"The Decoder：AI News","aggregationSource":"The Decoder：AI News","pageTitle":"LEGO-Anything araştırması: Kodlama ajanları tek bir fotoğrafı çalıştırılabilir bir Blender 3D sahne programına dönüştürebiliyor, ancak geometriye ilişkin öz değerlendirmeleri rastgeleye yakın. - Aioga AI Haberleri","description":"LEGO-Anything, kodlama ajanlarının tek bir görüntüden yola çıkarak yürütülebilir Blender sahne programlarını yinelemeli biçimde yazmasını sağlar ve simülasyonla oluşturulan görüntü...","url":"https://www.aioga.com/tr/news/fjxvgq4f2pnypa9f5or45o1i3/","contentTranslated":true,"translationStatus":"translated","translationRetryAt":"","translationError":"","sourceHash":"cb81ddf8f1934e86","translatedAt":"2026-10-03T19:00:06.109Z"},"vi":{"title":"Nghiên cứu LEGO-Anything: Tác nhân lập trình có thể chuyển một bức ảnh duy nhất thành chương trình cảnh 3D Blender có thể thực thi, nhưng khả năng tự đánh giá hình học gần như ngẫu nhiên","summary":"LEGO-Anything cho phép các tác nhân lập trình từ một hình ảnh duy nhất lặp đi lặp lại để viết các chương trình cảnh Blender có thể thực thi, đồng thời ra mắt chuẩn đo LEGO-Bench (208 hình ảnh, 104 cảnh, 443 tài sản đã đăng ký), cung cấp giá trị thực 3D chính xác thông qua mô phỏng render.","category":"行业动态","source":"The Decoder：AI News","aggregationSource":"The Decoder：AI News","pageTitle":"Nghiên cứu LEGO-Anything: Tác nhân lập trình có thể chuyển một bức ảnh duy nhất thành chương trình cảnh 3D Blender có thể thực thi, nhưng khả năng tự đánh giá hình học gần như ngẫu nhiên - Tin tức AI Aioga","description":"LEGO-Anything cho phép các tác nhân lập trình từ một hình ảnh duy nhất lặp đi lặp lại để viết các chương trình cảnh Blender có thể thực thi, đồng thời ra mắt chuẩn đo LEGO-Bench (2...","url":"https://www.aioga.com/vi/news/fjxvgq4f2pnypa9f5or45o1i3/","contentTranslated":true,"translationStatus":"translated","translationRetryAt":"","translationError":"","sourceHash":"cb81ddf8f1934e86","translatedAt":"2026-10-03T19:00:10.383Z"},"id":{"title":"Penelitian LEGO-Anything: Agen pengkodean cerdas dapat mengubah satu foto menjadi program adegan 3D Blender yang dapat dijalankan tetapi evaluasi geometrinya hampir acak","summary":"LEGO-Anything memungkinkan agen pemrograman untuk menulis program adegan Blender yang dapat dieksekusi dari satu gambar secara iteratif, dan meluncurkan tolok ukur LEGO-Bench (208 gambar, 104 adegan, 443 aset terdaftar), dengan rendering simulasi menyediakan nilai kebenaran 3D yang akurat.","category":"行业动态","source":"The Decoder：AI News","aggregationSource":"The Decoder：AI News","pageTitle":"Penelitian LEGO-Anything: Agen pengkodean cerdas dapat mengubah satu foto menjadi program adegan 3D Blender yang dapat dijalankan tetapi evaluasi geometrinya hampir acak - Berita AI Aioga","description":"LEGO-Anything memungkinkan agen pemrograman untuk menulis program adegan Blender yang dapat dieksekusi dari satu gambar secara iteratif, dan meluncurkan tolok ukur LEGO-Bench (208...","url":"https://www.aioga.com/id/news/fjxvgq4f2pnypa9f5or45o1i3/","contentTranslated":true,"translationStatus":"translated","translationRetryAt":"","translationError":"","sourceHash":"cb81ddf8f1934e86","translatedAt":"2026-10-03T19:00:35.422Z"},"th":{"title":"การวิจัย LEGO-Anything: ตัวแทนรหัสสามารถแปลงภาพถ่ายเพียงภาพเดียวให้เป็นโปรแกรมฉาก 3D ของ Blender ที่สามารถรันได้ แต่การประเมินรูปทรงใกล้เคียงกับการสุ่ม","summary":"LEGO-Anything ทำให้เอเจนต์การเข้ารหัสสามารถสร้างโปรแกรมฉาก Blender ที่ทำงานได้จากภาพเพียงภาพเดียวแบบวนซ้ำ และเปิดตัวมาตรฐาน LEGO-Bench (208 ภาพ, 104 ฉาก, 443 สินทรัพย์ที่ลงทะเบียน) โดยใช้การจำลองการเรนเดอร์เพื่อให้ค่าจริง 3D ที่แม่นยำ","category":"行业动态","source":"The Decoder：AI News","aggregationSource":"The Decoder：AI News","pageTitle":"การวิจัย LEGO-Anything: ตัวแทนรหัสสามารถแปลงภาพถ่ายเพียงภาพเดียวให้เป็นโปรแกรมฉาก 3D ของ Blender ที่สามารถรันได้ แต่การประเมินรูปทรงใกล้เคียงกับการสุ่ม - ข่าว AI Aioga","description":"LEGO-Anything ทำให้เอเจนต์การเข้ารหัสสามารถสร้างโปรแกรมฉาก Blender ที่ทำงานได้จากภาพเพียงภาพเดียวแบบวนซ้ำ และเปิดตัวมาตรฐาน LEGO-Bench (208 ภาพ, 104 ฉาก, 443 สินทรัพย์ที่ลงทะเบียน)...","url":"https://www.aioga.com/th/news/fjxvgq4f2pnypa9f5or45o1i3/","contentTranslated":true,"translationStatus":"translated","translationRetryAt":"","translationError":"","sourceHash":"cb81ddf8f1934e86","translatedAt":"2026-10-03T19:05:30.889Z"},"pl":{"title":"Badania LEGO-Anything: Agent kodujący może przekształcić pojedyncze zdjęcie w wykonalny program sceny 3D w Blenderze, ale samoocena geometrii jest bliska losowej","summary":"LEGO-Anything pozwala agentowi kodującemu iteracyjnie tworzyć wykonalne programy scen Blender z pojedynczego obrazu i wprowadza benchmark LEGO-Bench (208 obrazów, 104 sceny, 443 zarejestrowane zasoby), dostarczając dokładną prawdziwą wartość 3D dzięki symulacji renderowania.","category":"行业动态","source":"The Decoder：AI News","aggregationSource":"The Decoder：AI News","pageTitle":"Badania LEGO-Anything: Agent kodujący może przekształcić pojedyncze zdjęcie w wykonalny program sceny 3D w Blenderze, ale samoocena geometrii jest bliska losowej - Aioga Wiadomości AI","description":"LEGO-Anything pozwala agentowi kodującemu iteracyjnie tworzyć wykonalne programy scen Blender z pojedynczego obrazu i wprowadza benchmark LEGO-Bench (208 obrazów, 104 sceny, 443 za...","url":"https://www.aioga.com/pl/news/fjxvgq4f2pnypa9f5or45o1i3/","contentTranslated":true,"translationStatus":"translated","translationRetryAt":"","translationError":"","sourceHash":"cb81ddf8f1934e86","translatedAt":"2026-10-03T19:05:09.533Z"}},"evidenceTier":"verified-news","reviewStatus":"automated-ingest","indexable":true,"editorialCover":""}}