Makeup using Copycat
In the last class, Gonzalo assigned us to play with Copycat, which involved adding makeup to a provided video. To set the reference frames, I searched for an image featuring makeup. Then, I erased the surroundings, leaving only the makeup in the image. Afterward, I prepared six frames for training with eye makeup. By connecting the original clip to the input and the prepared clip to the groundtruth, I finished the preparation for training Copycat.
After the training, I obtained a .cat file and applied it to the original source. However, there was an error in the makeup area. To address this, I removed the Crop node in case it interfered with the training, and I increased the number of epochs.
This is the result with 30,000 epochs. While it’s not perfect, I believe that adding more frames or training with significantly higher epochs could yield a more natural result. At the moment, I don’t consider it usable for professional work, but it’s an intriguing concept that has the potential for further development.