萌学园

谷歌 DeepMind 发布 GenCeption：打破计算机视觉桎梏_我的网站

一 |     Samantha Ruth Prabhu and Raj Nidimoru entered wedlock a few days ago! While the newlyweds are still receiving congratulatory messages, we have learnt that the couple would collaborate professionally to bankroll films.          Samantha, Raj collaborate professionally     Much before their wedding, Samantha and Raj collaborated as director and actor, and also as a co-producer. Last year, Samantha launched her production house, Tralala Moving Pictures under which she produced the film, Subham. Raj was credited as creative producer for the film while one of his associates oversaw the production. Similarly, for Samantha’s upcoming film, Maa Inti Bangaram, Raj is the creative producer. But according to sources, Raj would be elevated to not just overseeing creative aspects but also involved in financial decisions, officially. Samantha worked with Raj for projects like The Family Man and Citadel: Honey Bunny. She knew his rich experience as a producer will come in handy in bringing down the cost of her production ventures. So going forward, we can see Raj would be credited as ‘producer’ for all Samantha’s films. On the other hand, Raj continues to work on his own web series in collaboration with DK (aka Krishna Dasarakothapalli). The duo reportedly has a couple of Hindi web shows lined up, apart from two theatrical releases.          Sam-Raj to continue work from Hyderabad     While Samantha and Raj have a residence in Mumbai, they continue to operate from Hyderabad where they also have a home. Both of them are set to return home shortly and resume work for their respective projects. Samantha is also in talks to headline a couple of Telugu films. It is being speculated that she is likely to announce them shortly. While balancing production (in collaboration with her husband Raj Nidimoru), Samantha aspires to act in films, although she is selective in picking up films.     Also Read: Samantha Ruth Prabhu Reacts to Sheetal Nidimoru’s Heartfelt Post on Wedding。

二 |     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。         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。         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。

三 |          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。

四 |          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。         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。

五 |          参考。

Current article:http://b1sm2.shaihanhuangwoshang.cyou/news/20260826_710.html

Published on:01:45:42