Open yjhboy opened 1 month ago
Thank you for your outstanding contributions, and I wonder if the target term for this line (https://github.com/EnVision-Research/LucidDreamer/blob/2ecf0936617103107e4b20c34e94d204196f7a44/guidance/sd_utils.py#L279) should theoretically be equal to unconditional unet output (pred_noise) of xt, ie., (\epsilon\phi(x_t,t,\emptyset)) ? Rather than be xs in the paper, ie., \epsilon\phi(x_s,s,\emptyset) ?
Thank you for your outstanding contributions, and I wonder if the target term for this line (https://github.com/EnVision-Research/LucidDreamer/blob/2ecf0936617103107e4b20c34e94d204196f7a44/guidance/sd_utils.py#L279) should theoretically be equal to unconditional unet output (pred_noise) of xt, ie., (\epsilon\phi(x_t,t,\emptyset)) ? Rather than be xs in the paper, ie., \epsilon\phi(x_s,s,\emptyset) ?