What Sketch Explainability Really Means for Downstream Tasks
This addresses the need for explainable AI in sketch-related applications, though it appears to be an incremental improvement by adapting existing explainability concepts to the sketch modality.
The paper tackles the problem of explainability for sketch-based AI models by proposing a lightweight plugin that provides stroke-level attribution maps without requiring model retraining. The solution demonstrates adaptability across four applications including retrieval, generation, assisted drawing, and sketch adversarial attacks.
In this paper, we explore the unique modality of sketch for explainability, emphasising the profound impact of human strokes compared to conventional pixel-oriented studies. Beyond explanations of network behavior, we discern the genuine implications of explainability across diverse downstream sketch-related tasks. We propose a lightweight and portable explainability solution -- a seamless plugin that integrates effortlessly with any pre-trained model, eliminating the need for re-training. Demonstrating its adaptability, we present four applications: highly studied retrieval and generation, and completely novel assisted drawing and sketch adversarial attacks. The centrepiece to our solution is a stroke-level attribution map that takes different forms when linked with downstream tasks. By addressing the inherent non-differentiability of rasterisation, we enable explanations at both coarse stroke level (SLA) and partial stroke level (P-SLA), each with its advantages for specific downstream tasks.