Where is My Spot? Few-shot Image Generation via Latent Subspace Optimization

Input (b) AGE [7] (c) WaveGAN [44] (d) Ours Figure 1 . We propose a new image synthesis approach that allows generating diverse unseen results with only 1-or 3-shot samples. Our key idea is to exploit the continuity of the StyleGAN latent space, and further empower it to generate unseen categories via latent subspace optimization.