Video Embedding for recommendation systems

[thomasf1]

Is there an implementation for this approach: https://milvus.io/blog/2021-10-10-milvus-helps-analyze-videos.md ?

Would be interesting to compare it to compare it to the performance of pytorchvideo/SlowFast.

Currently, generating embeddings for video seems to be a far way off what is possible for images.

Did anyone else work on this?

[yhmo]

An example application is here: https://milvus.io/docs/video_similarity_search.md

[thomasf1]

@yhmo Thanks, yhmo - The approach in the examples is different from the article, it´s using timesformer and pytorchvideo via towhee. I wanted to compare the two approaches... :)