FlowFace: Semantic Flow-guided Shape-aware Face Swapping

GrowingNet: An end-to-end growing network for semi-supervised learning

“…1) Quantitative Comparisons. : Our method is compared with seven methods including Deepfakes [22], FaceSwap [20], FlowFace [6], FSGAN [23], FaceShifter [2], SimSwap [3], and HifiFace [7]. For Deepfakes, FaceSwap, FaceShifter, and HifiFace, we use their released face swapping results of the sampled 10,000 images.…”

“…Our previous work, FlowFace [6], is a two-stage framework. It uses the D shape 's warped results on the target faces as the first stage, then utilizes the F swa which is trained without D shape 's supervision to transfer the non-shape identity as the second stage.…”

“…Recent works [2]- [6] have made great improvements in achieving promising face-swapping results. However, many of them focus on transferring inner facial features, while neglecting facial contour reshaping.…”

“…However, many of them focus on transferring inner facial features, while neglecting facial contour reshaping. We are aware that facial contours also play a crucial role in conveying a person's identity, yet few efforts [6], [7] have been devoted to exploring contour transferring. Reshaping facial contours presents significant challenges as it involves making substantial changes to the pixel values.…”

“…HifiFace [7] directly utilizes a 3D facial reconstruction model with the coarse perception of face shape. FlowFace [6] adopts a two-stage framework with a specific face-reshaping network and a face-swapping network. It suffers from error accumulation introduced by the two individual stages.…”