Layout Generation for Design

Module: Generative AI | Difficulty: Advanced

Room Layout Estimation

where are corners and are wall segments.

Graphic Design Layout

UI Generation

class LayoutGenerator(nn.Module):
    def __init__(self, n_classes=10, embed_dim=128):
        super().__init__()
        self.type_embed = nn.Embedding(n_classes, embed_dim)
        self.pos_mlp = nn.Sequential(
            nn.Linear(embed_dim, 256), nn.ReLU(),
            nn.Linear(256, 4))  # x, y, w, h
    def forward(self, n_elements, conditioning):
        types = torch.arange(n_elements).unsqueeze(0).expand(conditioning.size(0), -1)
        h = self.type_embed(types) + conditioning
        return self.pos_mlp(h).sigmoid()

Research Insight: Layout generation is fundamentally a set prediction problem. DETR-style architectures with bipartite matching losses achieve better results than autoregressive approaches for layout generation.