Slow-Fast Learning for Action-Conditioned Long Video Generation
Slow-Fast Learning for Action-Conditioned Long Video Generation
Additional Visualization for rebuttal
Embedded Files
Section1: Overview
Section 2: More Fast Learning Examples
Section 3: More Slow Learning Examples
Section 4: Examples on Action Generalization
Action Interpolation
Action Interpolation
Jog on
Simlar Action
Simlar Action
Sterring the bicycle to the right
Action Composition
Action Composition
Jump while moving left and forward
Section 5: Examples on Walking Tour
Section 6: Examples on Scene Revisit with Dynamic Objects
Section 7: Results on Teco-Habitat (Action-Conditioned)
The long videos are composed by 53/61 generated short video sequences respectively (there are 53/61 actions in the generated long video).
conditioned video
conditioned video
frame < 144
wo Temp-LoRA
wo Temp-LoRA
frame > 144
w Temp-LoRA
w Temp-LoRA
frame > 144
conditioned video
conditioned video
frame < 144
wo Temp-LoRA
wo Temp-LoRA
frame > 144
w Temp-LoRA
w Temp-LoRA
frame > 144
Page updated
Google Sites
Report abuse