All public logs

Combined display of all available logs of Marovi. You can narrow down the view by selecting a log type, the username (case-sensitive), or the affected page (also case-sensitive).

• 00:31, 9 September 2024 Felipefelixarias talk contribs created page Translations:Diffusion Models Are Real-Time Game Engines/95/zh (Created page with "== 贡献 ==")
• 00:30, 9 September 2024 Felipefelixarias talk contribs created page Translations:Diffusion Models Are Real-Time Game Engines/94/zh (Created page with "我们衷心感谢 Eyal Segalis、Eyal Molad、Matan Kalman、Nataniel Ruiz、Amir Hertz、Matan Cohen、Yossi Matias、Yael Pritch、Danny Lumen、Valerie Nygaard、Theta Labs 和 Google Research 团队以及我们的家人，感谢他们提供的深刻反馈、创意、建议和支持。")
• 00:30, 9 September 2024 Felipefelixarias talk contribs created page Translations:Diffusion Models Are Real-Time Game Engines/93/zh (Created page with "== 致谢 ==")
• 00:30, 9 September 2024 Felipefelixarias talk contribs created page Translations:Diffusion Models Are Real-Time Game Engines/92/zh (Created page with "'''迈向互动视频游戏的新范式''' 如今，视频游戏是由人类''编程''的。而''GameNGen''则是新范式的一部分概念验证，在这一新范式中，游戏被视为神经模型的权重，而不是代码行。''GameNGen''展示了一种架构和模型权重，使神经模型能够有效地在现有硬件上互动运行复杂的游戏（如 DOOM）。虽然仍有许多重要问题，但我们希望这种范式能带来显著益处。例如，在这...")
• 00:30, 9 September 2024 Felipefelixarias talk contribs created page Translations:Diffusion Models Are Real-Time Game Engines/91/zh (Created page with "'''未来工作''' 我们在经典游戏《DOOM》上演示了''GameNGen''。在其他游戏或更广泛的交互式软件系统上进行测试会非常有趣。我们注意到，除了强化学习代理的奖励函数外，我们的技术没有任何内容是《DOOM》特有的。我们计划在未来的工作中解决这个问题。虽然''GameNGen''能够准确地维护游戏状态，但如上文所述，它并不完美。可能需要一个更复杂的架构来缓解...")
• 00:30, 9 September 2024 Felipefelixarias talk contribs created page Translations:Diffusion Models Are Real-Time Game Engines/90/zh (Created page with "'''局限性。''' GameNGen 存在内存有限的情况。该模型只能访问稍超过3秒的历史记录，因此许多游戏逻辑能够在更长的时间跨度内被保存，这一点令人惊讶。虽然部分游戏状态是通过屏幕像素（如弹药和健康统计、可用武器等）来持久化的，但模型可能学习了强大的启发式方法，从而能够进行有意义的概括。例如，从渲染视图中，模型可以学习推断玩家的位置...")
• 00:30, 9 September 2024 Felipefelixarias talk contribs created page Translations:Diffusion Models Are Real-Time Game Engines/89/zh (Created page with "'''总结。''' 我们介绍了''GameNGen''，并证明在神经模型上可以实现每秒20帧的高质量实时游戏。我们还提供了将计算机游戏等交互式软件转换为神经模型的方法。")
• 00:30, 9 September 2024 Felipefelixarias talk contribs created page Translations:Diffusion Models Are Real-Time Game Engines/88/zh (Created page with "== 7 讨论 ==")
• 00:30, 9 September 2024 Felipefelixarias talk contribs created page Translations:Diffusion Models Are Real-Time Game Engines/87/zh (Created page with "DOOM 于 1993 年发布，掀起了游戏行业的一场革命。它引入了开创性的 3D 图形技术，成为第一人称射击类游戏的基石，影响了无数其他游戏。许多研究工作都对 DOOM 进行了研究。它提供了开放源码的实现和足够低的原生分辨率，适合小型模型的模拟，同时也足够复杂，可以作为一个具有挑战性的测试案例。最后，作者在这款游戏上花费了无数的青春时光，因此...")
• 00:30, 9 September 2024 Felipefelixarias talk contribs created page Translations:Diffusion Models Are Real-Time Game Engines/86/zh (Created page with "===== DOOM =====")
• 00:30, 9 September 2024 Felipefelixarias talk contribs created page Translations:Diffusion Models Are Real-Time Game Engines/85/zh (Created page with "有几项研究试图利用动作输入来训练游戏仿真模型。Yang 等人（[https://arxiv.org/html/2408.14837v1#bib.bib38 2023]）建立了一个包含真实世界和模拟视频的多样化数据集，并训练了一个扩散模型，根据前一个视频片段和动作的文字描述来预测后续视频。Menapace 等人（[https://arxiv.org/html/2408.14837v1#bib.bib18 2021]）和 Bruce 等人（[https://arxiv.org/html/2408.14837v1#bib.bib7 2024]）专注于...")
• 00:30, 9 September 2024 Felipefelixarias talk contribs created page Translations:Diffusion Models Are Real-Time Game Engines/84/zh (Created page with "===== 游戏模拟与世界模型 =====")
• 00:29, 9 September 2024 Felipefelixarias talk contribs created page Translations:Diffusion Models Are Real-Time Game Engines/83/zh (Created page with "扩散模型在文本到图像生成中取得了最先进的成果（Saharia 等人，[https://arxiv.org/html/2408.14837v1#bib.bib27 2022]；Rombach 等人，[https://arxiv.org/html/2408.14837v1#bib.bib26 2022]；Ramesh 等人，[https://arxiv.org/html/2408.14837v1#bib.bib25 2022]；Podell 等人，[https://arxiv.org/html/2408.14837v1#bib.bib23 2023]），这一研究领域也被应用于文本到视频生成任务（Ho 等人，[https://arxiv.org/html/2408.14837v1#bib.b...")
• 00:29, 9 September 2024 Felipefelixarias talk contribs created page Translations:Diffusion Models Are Real-Time Game Engines/82/zh (Created page with "===== 视频扩散模型 =====")
• 00:29, 9 September 2024 Felipefelixarias talk contribs created page Translations:Diffusion Models Are Real-Time Game Engines/81/zh (Created page with "重建三维表示的神经方法在过去几年中取得了重大进展。NeRFs（Mildenhall 等人，[https://arxiv.org/html/2408.14837v1#bib.bib20 2020]）使用深度神经网络对辐射场进行参数化，该网络针对从不同相机姿态拍摄的一组图像的特定场景进行了专门优化。训练完成后，可通过体积渲染方法对场景的新视角进行采样。Gaussian Splatting（Kerbl 等人，[https://arxiv.org/html/2408.14837v1#bib.bib15 202...")
• 00:29, 9 September 2024 Felipefelixarias talk contribs created page Translations:Diffusion Models Are Real-Time Game Engines/80/zh (Created page with "===== 神经三维仿真 =====")
• 00:29, 9 September 2024 Felipefelixarias talk contribs created page Translations:Diffusion Models Are Real-Time Game Engines/79/zh (Created page with "模拟二维和三维环境的视觉和物理过程，并允许对其进行交互式探索，是计算机图形学中一个广泛发展的领域（Akenine-Möller等人，[https://arxiv.org/html/2408.14837v1#bib.bib1 2018]）。像虚幻和Unity这样的游戏引擎是可以处理场景几何表示并根据用户交互渲染图像流的软件。游戏引擎负责跟踪所有世界状态，例如玩家的位置和移动、物体、角色动画和光照。它还负责跟...")
• 00:29, 9 September 2024 Felipefelixarias talk contribs created page Translations:Diffusion Models Are Real-Time Game Engines/78/zh (Created page with "===== 交互式三维仿真 =====")
• 00:28, 9 September 2024 Felipefelixarias talk contribs created page Translations:Diffusion Models Are Real-Time Game Engines/77/zh (Created page with "== 6 相关工作 ==")
• 00:28, 9 September 2024 Felipefelixarias talk contribs created page Translations:Diffusion Models Are Real-Time Game Engines/76/zh (Created page with "{| class="wikitable" ! 难度级别 ! 数据生成策略 ! PSNR <math>\uparrow</math> ! LPIPS <math>\downarrow</math> |- | 简单 | 代理 | <math>20.94 \pm 0.76</math> | <math>0.48 \pm 0.01</math> |- | | 随机 | <math>20.20 \pm 0.83</math> | <math>0.48 \pm 0.01</math> |- | 中等 | 代理 | <math>20.21 \pm 0.36</math> | <math>0.50 \pm 0.01</math> |- | | 随机 | <math>16.50 \pm 0.41</math> | <math>0.59 \pm 0.01</math> |- | 困难 | 代理 | <math>17.51 \pm 0.35</math...")
• 00:28, 9 September 2024 Felipefelixarias talk contribs created page Translations:Diffusion Models Are Real-Time Game Engines/75/zh (Created page with "'''表 2：不同难度级别的表现。''' 我们比较了使用代理生成数据和随机生成数据训练的模型在简单、中等和困难数据集上的表现。简单和中等数据集各有 112 个样本，困难数据集有 232 个样本。在 3 秒后的单帧上计算每个轨迹的指标。")
• 00:28, 9 September 2024 Felipefelixarias talk contribs created page Translations:Diffusion Models Are Real-Time Game Engines/74/zh (Created page with "总体而言，我们观察到在随机轨迹上训练模型的效果出奇地好，但受到随机策略探索能力的限制。在比较单帧生成时，代理的效果稍好，PSNR 为 25.06，而随机策略为 24.42。在比较 3 秒自回归生成后的帧时，差距增大到 19.02 对 16.84。在手动操作模型时，我们发现某些区域对两者都很容易，而某些区域对两者都很困难，而在某些区域，代理的表现要好得多。基于...")
• 00:28, 9 September 2024 Felipefelixarias talk contribs created page Translations:Diffusion Models Are Real-Time Game Engines/73/zh (Created page with "我们将代理生成的数据训练与使用随机策略生成的数据训练进行比较。对于随机策略，我们根据与观测结果无关的均匀分类分布对动作进行采样。我们通过对两个模型及其解码器进行")
• 00:28, 9 September 2024 Felipefelixarias talk contribs created page Translations:Diffusion Models Are Real-Time Game Engines/72/zh (Created page with "==== 5.2.3 代理执行 ====")
• 00:28, 9 September 2024 Felipefelixarias talk contribs created page Translations:Diffusion Models Are Real-Time Game Engines/71/zh (Created page with "center|thumb|600x600px|图 7：噪声增强的影响。图中显示了每个自回归步骤的 PSNR 平均值（越高越好）。不使用噪声增强时，质量在 10-20 帧后迅速下降。噪声增强可以防止这种情况。")
• 00:28, 9 September 2024 Felipefelixarias talk contribs created page Translations:Diffusion Models Are Real-Time Game Engines/70/zh (Created page with "center|thumb|600x600px|图 7：噪声增强的影响。图中显示了每个自回归步骤的 LPIPS 平均值（越低越好）。未使用噪声增强时，质量在 10-20 帧后迅速下降，而噪声增强可以防止这种情况。")
• 00:28, 9 September 2024 Felipefelixarias talk contribs created page Translations:Diffusion Models Are Real-Time Game Engines/69/zh (Created page with "在没有噪声增强的情况下，与真实值相比，LPIPS 距离迅速增加，而 PSNR 下降，这表明仿真结果与真实值的偏差加大。")
• 00:28, 9 September 2024 Felipefelixarias talk contribs created page Translations:Diffusion Models Are Real-Time Game Engines/68/zh (Created page with "为了消除噪声增强的影响，我们训练了一个不添加噪声的模型。我们对标准噪声增强模型和不添加噪声的模型（经过 200,000 步训练后）进行自回归评估，并计算在随机保留的 512 条轨迹上预测帧与真实帧之间的 PSNR 和 LPIPS 指标。我们在图 [https://arxiv.org/html/2408.14837v1#S5.F7 7] 中报告了每个自回归步骤的平均指标值，最多可达 64 帧。")
• 00:27, 9 September 2024 Felipefelixarias talk contribs created page Translations:Diffusion Models Are Real-Time Game Engines/67/zh (Created page with "==== 5.2.2 噪声增强 ====")
• 00:27, 9 September 2024 Felipefelixarias talk contribs created page Translations:Diffusion Models Are Real-Time Game Engines/66/zh (Created page with "{| class="wikitable" ! 历史上下文长度 ! PSNR <math>\uparrow</math> ! LPIPS <math>\downarrow</math> |- | 64 | <math>22.36 \pm 0.033</math> | <math>0.295 \pm 0.001</math> |- | 32 | <math>22.31 \pm 0.033</math> | <math>0.296 \pm 0.001</math> |- | 16 | <math>22.28 \pm 0.033</math> | <math>0.296 \pm 0.001</math> |- | 8 | <math>22.26 \pm 0.033</math> | <math>0.296 \pm 0.001</math> |- | 4 | <math>22.26 \pm 0.034</math> | <math>0.298 \pm 0.001</math> |- | 2 | <math>22.03...")
• 00:27, 9 September 2024 Felipefelixarias talk contribs created page Translations:Diffusion Models Are Real-Time Game Engines/65/zh (Created page with "表 1：历史帧数量。我们在来自 5 个级别的 8912 个测试集示例中分析了用作上下文的历史帧数量。更多的帧通常会改善 PSNR 和 LPIPS 指标。")
• 00:27, 9 September 2024 Felipefelixarias talk contribs created page Translations:Diffusion Models Are Real-Time Game Engines/64/zh (Created page with "我们通过训练使用<math>N \in \{ 1,2,4,8,16,32,64\}</math>的模型来评估改变条件上下文中过去观测值数量<math>N</math>的影响（请注意，我们的方法使用<math>N = 64</math>）。这影响了历史帧和动作的数量。我们在解码器保持冻结的情况下训练模型200,000步，并在5个级别的测试集轨迹上进行评估。结果见表[https://arxiv.org/html/2408.14837v1#S5.T1 1]。正如预期的那样，我们发现生成...")
• 00:27, 9 September 2024 Felipefelixarias talk contribs created page Translations:Diffusion Models Are Real-Time Game Engines/63/zh (Created page with "==== 5.2.1 上下文长度 ====")
• 00:27, 9 September 2024 Felipefelixarias talk contribs created page Translations:Diffusion Models Are Real-Time Game Engines/62/zh (Created page with "为了评估我们方法中不同组件的重要性，我们从评估数据集中采样轨迹，并计算真实值与预测帧之间的 LPIPS 和 PSNR 指标。")
• 00:26, 9 September 2024 Felipefelixarias talk contribs created page Translations:Diffusion Models Are Real-Time Game Engines/61/zh (Created page with "=== 5.2 消融实验 ===")
• 00:26, 9 September 2024 Felipefelixarias talk contribs created page Translations:Diffusion Models Are Real-Time Game Engines/60/zh (Created page with "'''人类评估。''' 作为评估仿真质量的另一项标准，我们向 10 名评测员提供了 130 个随机短片段（长度为 1.6 秒和 3.2 秒），并排展示我们的仿真和真实游戏。评测员的任务是识别真实游戏（见附录[https://arxiv.org/html/2408.14837v1#A1.SS6 A.6]中的图[https://arxiv.org/html/2408.14837v1#A1.F14 14]）。评测员在 1.6 秒和 3.2 秒的片段中，选择真实游戏而非仿真的比例分别为 58% 和 60%。")
• 00:26, 9 September 2024 Felipefelixarias talk contribs created page Translations:Diffusion Models Are Real-Time Game Engines/59/zh (Created page with "因此，我们对512个随机保留的轨迹计算FVD（Unterthiner等人，[https://arxiv.org/html/2408.14837v1#bib.bib35 2019]），测量预测轨迹分布与真实值轨迹分布之间的距离，仿真的长度为16帧（0.8秒）和32帧（1.6秒）。对于16帧，我们的模型获得的FVD为<math>114.02</math>。对于32帧，我们的模型获得的FVD为<math>186.23</math>。")
• 00:26, 9 September 2024 Felipefelixarias talk contribs created page Translations:Diffusion Models Are Real-Time Game Engines/58/zh (Created page with "center|thumb|600x600px|图 6：自回归评估。64 个自回归步骤的 LPIPS 指标")
• 00:26, 9 September 2024 Felipefelixarias talk contribs created page Translations:Diffusion Models Are Real-Time Game Engines/57/zh (Created page with "center|thumb|600x600px|图 6：自回归评估。64步自回归过程中的PSNR指标。")
• 00:26, 9 September 2024 Felipefelixarias talk contribs created page Translations:Diffusion Models Are Real-Time Game Engines/56/zh (Created page with "视频质量 我们使用第[https://arxiv.org/html/2408.14837v1#S2 2]节中描述的自回归设置，按照真实轨迹所定义的动作序列对帧进行迭代采样，同时将模型自身的过往预测作为条件。自回归采样时，预测轨迹和真实轨迹常常在几步后发生偏离，这主要是由于不同轨迹的帧间积累了少量不同的运动速度。因此，如图[https://arxiv.org/html/2408.14837v1#S5.F6 6]所示，每帧的PSNR和LPIPS值...")
• 00:26, 9 September 2024 Felipefelixarias talk contribs created page Translations:Diffusion Models Are Real-Time Game Engines/55/zh (Created page with "center|thumb|900x900px|图 5：模型预测与地面实况对比。仅显示过去观测上下文的最后 4 帧。")
• 00:26, 9 September 2024 Felipefelixarias talk contribs created page Translations:Diffusion Models Are Real-Time Game Engines/54/zh (Created page with "'''图像质量。''' 我们使用第[https://arxiv.org/html/2408.14837v1#S2 2]节中描述的教师强迫设置来测量LPIPS（Zhang 等人，[https://arxiv.org/html/2408.14837v1#bib.bib40 2018]）和PSNR。在该设置中，我们对初始状态进行采样，并根据地面实况的过去观察轨迹预测单帧。在对5个不同级别的2048条随机轨迹进行评估时，我们的模型实现了<math>29.43</math>的PSNR值和<math>0.249</math>的LPIPS值。PSNR...")
• 00:25, 9 September 2024 Felipefelixarias talk contribs created page Translations:Diffusion Models Are Real-Time Game Engines/53/zh (Created page with "总体而言，从图像质量来看，我们的方法在长轨迹上实现了与原始游戏相当的仿真质量。对于短轨迹，人类评估者在区分仿真片段和实际游戏片段时，仅比随机猜测略胜一筹。")
• 00:25, 9 September 2024 Felipefelixarias talk contribs created page Translations:Diffusion Models Are Real-Time Game Engines/52/zh (Created page with "=== 5.1 仿真质量 ===")
• 00:25, 9 September 2024 Felipefelixarias talk contribs created page Translations:Diffusion Models Are Real-Time Game Engines/51/zh (Created page with "== 5 结果 ==")
• 00:25, 9 September 2024 Felipefelixarias talk contribs created page Translations:Diffusion Models Are Real-Time Game Engines/50/zh (Created page with "我们使用 Stable Diffusion 1.4 的预训练检查点训练所有仿真模型，解冻所有 U-Net 参数。我们使用的批量大小为 128，恒定学习率为 2e-5，采用无权重衰减的 Adafactor 优化器（Shazeer & Stern，[https://arxiv.org/html/2408.14837v1#bib.bib31 2018]），以及梯度剪切为 1.0。我们将扩散损失参数化更改为 v预测（Salimans & Ho [https://arxiv.org/html/2408.14837v1#bib.bib28 2022a]）。我们以 0.1 的概率去...")
• 00:25, 9 September 2024 Felipefelixarias talk contribs created page Translations:Diffusion Models Are Real-Time Game Engines/49/zh (Created page with "=== 4.2 生成模型训练 ===")
• 00:25, 9 September 2024 Felipefelixarias talk contribs created page Translations:Diffusion Models Are Real-Time Game Engines/48/zh (Created page with "代理模型使用 PPO（Schulman 等人，[https://arxiv.org/html/2408.14837v1#bib.bib30 2017]）进行训练，采用简单的 CNN 作为特征网络，基于 Mnih 等人（[https://arxiv.org/html/2408.14837v1#bib.bib21 2015]）的方法。在 CPU 上使用 Stable Baselines 3 基础架构（Raffin 等人，[https://arxiv.org/html/2408.14837v1#bib.bib24 2021]）进行训练。代理接收缩小后的帧图像和游戏地图，每个分辨率为 160x120。代理还可以...")
• 00:25, 9 September 2024 Felipefelixarias talk contribs created page Translations:Diffusion Models Are Real-Time Game Engines/47/zh (Created page with "=== 4.1 代理训练 ===")
• 00:24, 9 September 2024 Felipefelixarias talk contribs created page Translations:Diffusion Models Are Real-Time Game Engines/46/zh (Created page with "== 4 实验设置 ==")