I saw a dancing Paddington Bear in your project page, and I found that it was generated frame-by-frame.
I'm wondering if I want to try to let the model to see the "temporal" context of the video and then do the generation, then do I need to collect a bunch of "dancing Paddington Bear" and retrain the model? Because I found that in the code the model can only be trained on one class of objects at a time.I think if I want to train temporal context, the data that provides timing information should be of the same type of objects.
If my understanding above is correct, I was wondering if you could provide more information on where I can find these videos. I tried searching on objaverse, and I found that it does have a lot of animations, but they are scattered among various types of objects. There don't seem to be many animations for a certain category.
Thank you for your amazing work.
I saw a dancing Paddington Bear in your project page, and I found that it was generated frame-by-frame.
I'm wondering if I want to try to let the model to see the "temporal" context of the video and then do the generation, then do I need to collect a bunch of "dancing Paddington Bear" and retrain the model? Because I found that in the code the model can only be trained on one class of objects at a time.I think if I want to train temporal context, the data that provides timing information should be of the same type of objects.
If my understanding above is correct, I was wondering if you could provide more information on where I can find these videos. I tried searching on objaverse, and I found that it does have a lot of animations, but they are scattered among various types of objects. There don't seem to be many animations for a certain category.