📄 中文摘要
MCVD是一种新方法，能够预测视频片段的未来发展、填补缺失部分或生成全新的短视频。该技术通过学习隐藏部分帧的视频，来实现对未来或过去的预测。MCVD模型在处理视频时，采用逐帧的方式，使其能够逐步生成较长的视频。与复杂的记忆技巧不同，MCVD使用简单的图像构建步骤，便于运行。其生成的结果通常具有意想不到的清晰度和真实感，且适用于多种场景，展示了其灵活性和广泛应用潜力。
MCVD：用于预测、生成和插值的掩蔽条件视频扩散
📄 English Summary
MCVD: Masked Conditional Video Diffusion for Prediction, Generation, andInterpolation
MCVD is a novel approach that predicts future developments in video clips, fills in missing parts, or creates entirely new short videos. This technology learns from videos with some frames hidden, enabling it to predict the future or the past. The MCVD model processes videos in chunks of frames, allowing for step-by-step generation of longer videos. Unlike complex memory tricks, it employs simple image-building steps, making it easier to run. The results are often surprisingly sharp and realistic, demonstrating flexibility and potential for various types of scenes.