This paper proposes the first GAN inversion-based method for multi-class sketch-based
image generation (MC-SBIG). MC-SBIG is a challenging task that requires strong prior
knowledge due to the significant domain gap between sketches and natural images.
Existing learning-based approaches rely on a large-scale paired dataset to learn the
mapping between these two image modalities. However, since the public paired sketch-
photo data are scarce, it is struggling for learning-based methods to achieve satisfactory …

Conditional Encoder E. In Fig. 1, we show the architecture of our conditional encoder E. The
conditional encoder E takes a sketch s with a class label y as the input and outputs a latent
code z. It consists of five residual blocks with bottleneck layers [3], one convolutional layer,
one downsampling (max pooling) layer, and one linear projection layer. We use a shared
class embedding as the condition. As in [1], the condition vector of each block is linearly
projected to produce per-sample gains and biases for the Batch-Norm layers. The bias …