Thank you for your excellent work; I found it very interesting. I ran your process to perform cross-reenactment on a new input image. The results are good; however, since we only used one input image, the person's appearance in the output video differs from the original. To obtain more realistic results, should I train the process again using only a video of that specific person? Will this new training process lose all the knowledge gained from training the model with your dataset? Should I freeze some layers of the networks?
Thank you for your excellent work; I found it very interesting. I ran your process to perform cross-reenactment on a new input image. The results are good; however, since we only used one input image, the person's appearance in the output video differs from the original. To obtain more realistic results, should I train the process again using only a video of that specific person? Will this new training process lose all the knowledge gained from training the model with your dataset? Should I freeze some layers of the networks?