【SUPPLEMENTARY MATERIALS】MIGRATING FACE SWAP TO MOBILE DEVICES: A LIGHT-WEIGHT FRAMEWORK AND A SUPERVISED TRAINING SOLUTION
I. Introduction
[Paper] > [Code]
The supplementary materials include the detailed design of the network architecture, as well as the ablation experiments of the network architecture and more qualitative results, which are worth reading.
1. DESIGN OF THE FRAMEWORK
Based on our intensive experiments, we designed a lightweight GAN according to the following principles:
- Although increasing the number of channels can enrich the representation ability of feature maps, it will lead to an exponential increase in the number of parameters. To keep the network as lightweight as possible, we use a fixed number of channels in each layer. we pass