nickyfot / emoclip Goto Github PK

Hello,

I hope you're doing well. I recently came across your work and I find it truly intriguing. I was wondering how it differentiates from the research presented at BMVC 2023, which can be found here.

Specifically, I'm curious about the sample-level description generation technique you've introduced. The paper doesn't seem to provide an intuitive explanation regarding this. Is it possible for GPT to generate a coherent description directly from a video segment? Could you shed some light on how the sample-level description was generated? I'd greatly appreciate any insights you can share.

Thank you in advance for your time and clarification!

Best regards

Hi there,

I wrote the following on a closed issue, however I'm submitting a new issue for visibility.

I've downloaded the task weights and attempted to load the state dictionary into the model, but I encountered some missing keys related to the temporal transformer. A gist of a notebook can be found here

I really appreciate any help you may be able to offer in resolving this. I am eager to experiment with EmoCLIP :)

• Seth