As I dive into the world of video search, I’m faced with a daunting task: finding the perfect CLIP-like model. My goal is to implement semantic video search for my open-source data management project, Anagnorisis, to create a local YouTube-like experience. But, I’m struggling to find the right model.
I’ve been scouring the web, but all I can find is an outdated model from two years ago, VideoCLIP, which lacks licensing. Other models, like VJEP-A2-ViT-L/FP-C64-256, don’t provide text-aligned embeddings by default, and fine-tuning them would require a lot of effort and resources I don’t have.
I’ve even considered using screenshots with CLIP + audio embeddings to estimate the proper video-CLIP model, but that’s a last resort. I’m convinced there must be better alternatives out there.
So, I turned to the internet for help. Unfortunately, my Google searches and AI searches haven’t yielded any satisfying results. That’s why I’m reaching out to the community for help.
Do you know of any good alternatives? Are there other approaches I should consider? I’d love to hear from you.
## What’s at Stake
The ability to search videos by text is crucial for a seamless user experience. It’s what sets apart a good video search engine from a mediocre one. With the right CLIP-like model, I can create a system that understands the content of videos and retrieves relevant results based on user queries.
## The Future of Video Search
I’m not alone in this quest. The future of video search depends on our ability to develop models that can accurately understand and retrieve video content. It’s an area that’s still largely unexplored, and I’m excited to be a part of it.
If you have any insights or suggestions, please share them with me. Let’s work together to create a better video search experience.
—
*Check out my project, Anagnorisis, on GitHub: https://github.com/volotat/Anagnorisis*