DoubleXING

icon indicating a website link https://doubiiu.github.io/

icon company The Chinese University of Hong Kong icon location Hong Kong icon location Ph.D. student at CUHK

Results 100 comments of DoubleXING

Fine tune on longer video dataset

Please see #99.

Do you have any plans to release the interpolation training code?

Hi. I will release the training code this month. Currently, I am too busy with other research projects. Thanks for your interest and patience.

Do you have any plans to release the interpolation training code?

Done.

fine tune 256x256 model

We don't recommend finetuning the 256x256 model, since its performance is much, much, much worse than that of the 512x320 and 1024x576. models..

annotations of the constructed dataset

Hi. We will release it soon, as I am currently busy with other research projects.

annotations of the constructed dataset

Available now.

annotations of the constructed dataset

They are filtered as described in the paper.

annotations of the constructed dataset

Here I just use the ChatGPT to generate some keywords about the CGI/Graphics content and then filter those videos using caption only. A better way is to filter using image...

annotations of the constructed dataset

yeap, using CLIP following previous video generation works

How Dynamicrafter achieve frame interpolation?

For the image-to-video/image animation application, we just feed a single image to the i2v model, while for interpolation, we feed both the starting image and ending image to the model...