CVPR2024

SketchINR: A First Look into Sketches as Implicit Neural Representations

Hmrishav Bandyopadhyay, Ayan Kumar Bhunia, Pinaki Nath Chowdhury, Aneeshan Sain, Tao Xiang, Timothy M. Hospedales, Yi-Zhe Song

Abstract

Compressed Sketch Storage (Bytes) Figure 1 . SketchINR is an implicit neural representation for sequential vector sketches. It is the first neural representation with sufficient fidelity to be a drop-in replacement for the raw sketch data (a). The representation is significantly higher fidelity than existing learned representations such as SketchRNN [28], especially for more complex sketches (b, left). SketchINR also provides state-of-the-art sketch compression, with a substantially more compact representation than either vector or raster sketches (b, right).