Hi Team,
I am trying to generate a synthetic data (video) where only environment should be changed. I am using cosmos-transfer2.5. I have multiple format videos like rbg, segmented, edge and depth. These are used as input. I have added a text input mentioning the scene along with the changes in the environment. I am giving control parameters as edge at 1 and segmentation at 0.4-0.8 and depth at 0.6-1. But the output is varying very much. It is not retaining the ground truth as seen on the rgb video.
How to achieve this? I want everything that is seen on rgb video as same only the weather/environment to be changed.