A benchmark for rough sketch cleanup

Abstract

Our evaluation identifies shortcomings among state-of-the-art cleanup algorithms and discusses open problems for future research. CCS Concepts: • Computing methodologies → Image processing; Graphics systems and interfaces; Shape analysis; • Applied computing → Computer-aided design.

Color by Numbers: Interactive Structuring and Vectorization of Sketch Imagery

“…Complementary and concurrent to our research, Yan et al [32] confrm the premise that rough sketch cleanup and vectorization is an ambiguous and difcult problem, motivating interaction to communicate user intent. While they do not present an interactive solution, they provide a benchmark of diverse raster images in 5 categories (art, logo, architecture, fashion, product), along with artist hand-created clean-up, structuring, and vectorization (Figures 15, 16).…”

“…In the current world context, with communication increasingly shifting online, such sketches are scanned, and sometimes redrawn using vectors, so they may be neatened, textured, replicated and even animated. We present a novel interactive approach to this problem of cleanup, neatening, structuring and vectorization of rough raster sketches [32] (Figure 1). Increasingly, interfaces for creative expression are being designed to target simple, playful interaction, by non-experts, using simple mouse/trackpad interaction [27].…”

“…Vectorization of raster sketches is an important ongoing research problem [32]. There are two aspects to vectorization: recovering vector curves [8]; and structuring these curves [5], that are typically handled in sequence.…”

“…Our evaluation is threefold: we compare our workfow to automatic state-of-the-art vectorization algorithms, followed by vector drawing clean-up and region structuring using a commercial vector editing tool Inkscape [13]; we conduct an informal study of our approach with 12 novice users who were all able to use the tool efectively; and we show a diverse design gallery from a rough raster sketch benchmark [32], vectorized using Color-by-Numbers.…”

SketchZooms: Deep Multi‐view Descriptors for Matching Line Drawings

“…We selected pairs of sketches from the Yan et al. [YVG20] public dataset and computed dense correspondences. Figure 7 shows exemplary outputs of corresponding points in cartoons, manga, fashion and architectural sketches.…”