# Generating Sketches in a Hierarchical Auto-Regressive Process for Flexible Sketch Drawing Manipulation at Stroke-Level

Source: [Generating Sketches in a Hierarchical Auto-Regressive Process for Flexible Sketch Drawing Manipulation at Stroke-Level](<https://www.sota2.com/research/paper/691414742828b7cc6937e999>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-03-24T05:00:03.763Z

Authors: Sicong Zang, Shuhui Gao, Zhijun Fang

Published: 2025-11-11T06:40:12.000Z

## Abstract

Generating sketches with specific patterns as expected, i.e., manipulating sketches in a controllable way, is a popular task. Recent studies control sketch features at stroke-level by editing values of stroke embeddings as conditions. However, in order to provide generator a global view about what a sketch is going to be drawn, all these edited conditions should be collected and fed into generator simultaneously before generation starts, i.e., no further manipulation is allowed during sketch generating process. In order to realize sketch drawing manipulation more flexibly, we propose a hierarchical auto-regressive sketch generating process. Instead of generating an entire sketch at once, each stroke in a sketch is generated in a three-staged hierarchy: 1) predicting a stroke embedding to represent which stroke is going to be drawn, and 2) anchoring the predicted stroke on the canvas, and 3) translating the embedding to a sequence of drawing actions to form the full sketch. Moreover, the stroke prediction, anchoring and translation are proceeded auto-regressively, i.e., both the recently generated strokes and their positions are considered to predict the current one, guiding model to produce an appropriate stroke at a suitable position to benefit the full sketch generation. It is flexible to manipulate stroke-level sketch drawing at any time during generation by adjusting the exposed editable stroke embeddings.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Stroke Erasion on QuickDraw Human Evaluation](<https://www.sota2.com/research/sota/stroke-erasion-on-quickdraw-human-evaluation.md>): Stroke Erasion; QuickDraw Human Evaluation; Human Score: 2.3.

- [Stroke Replacement on QuickDraw Human Evaluation](<https://www.sota2.com/research/sota/stroke-replacement-on-quickdraw-human-evaluation.md>): Stroke Replacement; QuickDraw Human Evaluation; Human Score: 2.09.

- [Sketch Reconstruction on DS2](<https://www.sota2.com/research/sota/sketch-reconstruction-on-ds2.md>): Sketch Reconstruction; DS2; Reconstruction Score: 87.71.

- [Sketch Reconstruction on DS1](<https://www.sota2.com/research/sota/sketch-reconstruction-on-ds1.md>): Sketch Reconstruction; DS1; Rec: 89.9.

- [stroke-level sketch edit on DS2 QuickDraw (test)](<https://www.sota2.com/research/sota/stroke-level-sketch-edit-on-ds2-quickdraw-test.md>): stroke-level sketch edit; DS2 QuickDraw (test); Reconstruction: 63.72.

- [Stroke-Level Sketch Edit on DS1 QuickDraw (test)](<https://www.sota2.com/research/sota/stroke-level-sketch-edit-on-ds1-quickdraw-test.md>): stroke-level sketch edit; DS1 QuickDraw (test); Reconstruction: 49.25.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2511.07889>)

[PDF](<https://arxiv.org/pdf/2511.07889.pdf>)
