Offline virtual advertisement replacement in sports from uncalibrated video**

在这个项目中，我们将执行离线或突出体育广告替换纯粹从视频。在商业内容中进行这样的替换的原因可能是专注于更紧密地针对观众的节目。新的广告可以更适合观众的利益，它可以提高商业内容对发行商的价值。挑战在于，如此多的商业视觉内容与常规视频相似，单纯从视频中检测广告内容真的很难。此外，不可能存在关于原始广播广告的足够信息，并且它们的插入时隙通常不可用。在这一点上，为了克服所提到的挑战，该想法是应用诸如考虑基于梯度的优化的离线视觉广告检测的自动编码器之类的无监督方法。

In this project, we are going to perform offline or highlight sport advertisement replacement purely from video. The reason for such a replacement within commercial contents could be focusing on the programs which are more tightly targeted to the viewers. The new advertisement can be more suitable to the benefits of the viewers and it can enhance the value of the commercial content for the distributers. The challenge is that so many commercial visual contents are similar to the regular video and it would be really hard to detect the advertisement content purely from the video. Moreover, there would not possibly be enough information regarding the original broadcast advertisement and their insertion time slots are not usually available. In this regard, in order to overcome the mentioned challenge, the idea is to apply unsupervised methodologies such as autoencoder regarding offline visual advertisement detection considering gradient-based optimization.