Hyeonho, Jeong

icon indicating a website link https://hyeonho99.github.io/

icon company Adobe Research icon location Researching video generative AI at Adobe

Results 6 comments of Hyeonho, Jeong

This needs controlnet pose estimation

@Zebraslive I am actually working on that and will share the open source code when it's done.

unet/showone.models.unet_3d_condition.py as defined in `model_index.json` does not exist in showlab/show-1-base and is not a module in 'diffusers/pipelines'.

UNet3DConditionModel from _showone/modes/unet_3d_condition.py_ is used in "TextToVideoIFPipeline", which is locally defined in _showone/pipelines/pipeline_t2v_base_pixel.py_, which is also the initial keyframe generation model of the Show-1 cascaded vdm. Or did I understand...

In train.py, what is the usage/meaning of 'prior_preservation'?

This is not something included in the original paper right? Can you explain about this modification, please? Thank you.

In train.py, what is the usage/meaning of 'prior_preservation'?

Thank you for clarifying it. I will check out the Dreambooth paper. I have one more question. I think "train_temporal_conv" was not in the original Tune A Video paper. Is...

prompts for other examples

Hi, @gigoug - For **skiing**, the prompts used are these: - training prompt: "a man is snowboarding" - target prompts: "a spider-man is snowboarding", "an astronaut is snowboarding underwater in...

about the running time cost

Hi @CJ416 , Inference of a whole cascaded video diffusion pipeline takes some time. In the case of training with a single video of 8 frames, training should be done...