五感图

Depp Set to Return to Pirates of the Caribbean After Defamation Trial Win Against Heard - Report_我的网站

一 |     IT之家 7 月 21 日消息，科技媒体 The Decoder 昨日（7 月 20 日）发布博文，报道称谷歌 DeepMind 发布 GenCeption 模型，将预训练的视频生成器重新用于深度估计和分割等经典计算机视觉任务。         IT之家援引博文介绍，大语言模型在学习预测下一个 Token 的时候，在训练过程中往往需要吸收语法、世界知识和上下文关系等内容。         但是在计算机视觉领域，视觉模型缺少等效的训练方法，主要由专业模型主导，包括用于分割的“Segment Anything”和用于深度估计的“Depth Anything”，每个模型都使用其特定的架构。

二 |          谷歌 DeepMind 团队为此提出 GenCeption 模型方案，尝试将一个“生成视频”的 AI 模型逆向改造成一个能“理解世界”的视觉分析引擎。    Johnny Depp is in talks with Disney about his return to the Pirates of the Caribbean franchise of which he has been the star for more than 15 years, Poptopic reported, citing a "trusted source close to Disney".,According to the report, Disney is preparing a "$301Mln deal" by way of an apology for having dropped Depp from the business - where he has played iconic pirate, Jack Sparrow - in 2018 after his ex-wife Amber Heard's op-ed in which she described herself as "a public figure representing domestic abuse”.,The source told Poptopic that Disney "reached out to the actor prior to his defamation trial against Amber Heard". There is also another project awaiting Depp - the Disney+ series about the early life of the captain of the Black Pearl, the source added.,Amber Heard Claims Johnny Depp's Ex-Girlfriends 'Too Scared' to Accuse Him of Violence19 June, 12:53 GMT,The news comes just weeks after Depp won a defamation trial against Heard. The actor sued his ex-wife over an article in the Washington Post in 2018, in which she alleged she was a "public figure representing domestic violence".,Although Heard did not mention the 58-year-old Pirates of the Caribbean star, Depp insisted that her article harmed his reputation and career as he was dismissed from several films because of her claims.,Heard countersued Depp for $100Mln, claiming that her ex-husband and his former attorney Adam Waldman conspired to "defame" her. On 1 June, the jury ruled that both Depp and Heard were liable for defamation. However, the actor was awarded $15Mln in compensatory damages, and Heard received only $2Mln.。         GenCeption 打破了传统计算机视觉“一个任务一个专用模型”的格局，仅凭单一模型就能同时做好深度估计、图像分割、3D 姿态估计、表面法线预测和相机姿态估计等核心视觉任务。         GenCeption 基于阿里巴巴开源视频模型通义万相 Wan2.1 系列训练，与传统扩散模型需多步去噪不同，GenCeption 在一次前向传播中完成预测，从而提升视觉任务处理速度。模型通过文本提示指定任务，可输出深度图、表面法线图、分割掩码，并可处理相机运动表示。         训练数据以合成为主。论文称，数据集仅包含 7500 段视频，由 800 个数字人体模型与 200 段动作捕捉序列组合生成，再通过 Blender 在不同背景和镜头角度下渲染。

三 |          泛化方面，GenCeption 几乎只在单人合成视频上训练，但可处理真实多人视频，也可迁移到动物和类人机器人类别。论文称，部分输出细节甚至超过训练时 Blender 渲染结果，可保留猫胡须和单根发丝边缘。         性能方面，论文给出两组处理时间数据：小模型处理 81 帧视频约需 6 秒；大模型参数量为 140 亿，处理同样长度视频约需 10 秒。         参考。

Current article:http://ki210.minpinpengcemiuanjiuxuan.pics/9pe/hc0.html

Published on:05:39:33