This can happen sometimes, and the quality can be increased. Here is how:
- Make sure you start with good images. If the training A or training B images are blurry, the final output will be bad. Also, make sure you have at least a few hundreds of them.
- Add images with different lighting. If a light is on the left side, there will be shadows on the right side of the face. If you have a different photo with a light on the right side, the shadows will be on the left side of the face. This will help the A.I. algorithm to create a 3D map of the face.
- Train for longer. Make sure your Loss A and Loss B are around 0.02, 0.01. When the loss is not going down anymore, it means your A.I. can’t learn anymore for the existing data. You can add more photos, remove bad ones, or change some of the settings (increase the nodes or layers – make sure you backup your model first).
- Accept defeat. Sometimes, you can’t improve the result, even with the above suggestions. Take your experience so far with your failed project and apply it to your next project.