TactileEval: A Step Towards Automated Fine-Grained Evaluation and Editing of Tactile Graphics

arXiv cs.CV / 4/23/2026

📰 NewsDeveloper Stack & InfrastructureTools & Practical UsageModels & Research

共有:

Key Points

TactileEval proposes a three-stage pipeline to automate fine-grained evaluation and repair of tactile graphics, addressing the lack of actionable signals in existing datasets that only provide coarse quality scores.
The work builds a five-category quality taxonomy (view angle, part completeness, background clutter, texture separation, and line quality) derived from expert free-text feedback and aligned with BANA standards.
It collects 14,095 structured annotations from 66 object classes across six object families using Amazon Mechanical Turk, enabling both evaluation and editing workflows.
A reproducible ViT-L/14 feature probe trained on this data reaches 85.70% overall test accuracy across 30 tasks and shows consistent difficulty ordering that suggests the taxonomy reflects meaningful perceptual structure.
Using the evaluations, the authors introduce a ViT-guided automated editing pipeline that leverages family-specific prompt templates and gpt-image-1 image editing to generate targeted corrections.

Abstract

Tactile graphics require careful expert validation before reaching blind and visually impaired (BVI) learners, yet existing datasets provide only coarse holistic quality ratings that offer no actionable repair signal. We present TactileEval, a three-stage pipeline that takes a first step toward automating this process. Drawing on expert free-text comments from the TactileNet dataset, we establish a five-category quality taxonomy; encompassing view angle, part completeness, background clutter, texture separation, and line quality aligned with BANA standards. We subsequently gathered 14,095 structured annotations via Amazon Mechanical Turk, spanning 66 object classes organized into six distinct families. A reproducible ViT-L/14 feature probe trained on this data achieves 85.70% overall test accuracy across 30 different tasks, with consistent difficulty ordering suggesting the taxonomy suggesting the taxonomy captures meaningful perceptual structure. Building on these evaluations, we present a ViT-guided automated editing pipeline that routes classifier scores through family-specific prompt templates to produce targeted corrections via gpt-image-1 image editing. Code, data, and models are available at https://TactileEval.github.io/