TextEraseBench

TextEraseBench is a text-removal benchmark for evaluating object removal and inpainting methods on text overlays and scene text-like objects.

Dataset page: https://huggingface.co/datasets/QinmingZhou/TextEraseBench

Dataset Summary

TextEraseBench contains 185 paired samples. Each sample includes an input image containing target text, a paired clean background, and a removal mask. The benchmark is designed to test whether methods can remove text while preserving the surrounding background structure.

Dataset Structure

TextEraseBench/
├── shot/   # text-present input images
├── bg/     # paired clean background images
└── mask/   # target text/effect masks

For each sample id <stem>:

shot/<stem>.png
bg/<stem>.png
mask/<stem>.png

Usage

Use shot as the input image, mask as the removal condition, and bg as the paired target for evaluation.

Dataset Creation

TextEraseBench is constructed through a manual-to-automated pipeline. Target text regions are manually annotated with fine-grained bounding boxes, then removed to create paired clean backgrounds. Samples undergo secondary verification to filter artifacts and semantic inconsistencies.

Intended Use

TextEraseBench is intended for evaluating text removal, object removal, and image inpainting models on paired-background examples.