Sketch Representation

Sketch representation research focuses on developing effective computational methods to capture the essence of hand-drawn sketches, enabling tasks like sketch recognition, generation, and image synthesis. Current research emphasizes leveraging powerful pre-trained models like CLIP, autoregressive architectures (e.g., sequence-to-sequence models), and graph neural networks to represent sketches as sequences of primitives, graphs of interconnected patches, or stroke-based representations. These advancements improve the robustness and generalization capabilities of sketch-based systems, impacting applications ranging from human-computer interaction to robotics and creative design tools.