how to control motion magnitude

[hw-liang]

In the train.py file, I noticed for motion control, the motion magnitudes are computed in both RGB(batch["motion"]) and latent space (latent_motion = calculate_latent_motion_score(latents)), but only the latter is used in unet prediction. Could you explain why not use the former one but only use the latter one?

[sculmh]

In consideration of efficiency, if we opt to employ motion representation in the RGB format, it will be required to transform predict_x0 back into the RGB space in the line of code motion_loss = F.mse_loss(latent_motion, calculate_latent_motion_score(predict_x0)).