ViTeX-Bench: Benchmarking High-Fidelity Video Scene Text Editing
ViTeX-Bench and open ViTeX-Edit-14B target high-fidelity, temporally consistent video scene text editing.
ViTeX-Bench is a benchmark for replacing scene text in video while preserving motion, camera dynamics, and surrounding content. It includes 387 real-world 720p videos with masks and instructions: 230 reviewed paired edits for training and 157 held out for evaluation. A three-axis protocol uses 13 metrics covering text correctness, visual and temporal quality, and edit locality. The released ViTeX-Edit-14B editor, fine-tuned with motion-aligned glyph-video conditioning, scores CharAcc 0.688, the highest mean among evaluated video-native editors.
- ViTeX-Dataset has 387 real 720p videos, 230 paired training edits and 157 frozen eval clips.
- Thirteen metrics score text correctness, visual-temporal quality, and edit locality.
- Open ViTeX-Edit-14B reaches CharAcc 0.688, best among evaluated video-native editors.
- Eight baselines still struggle to combine accurate text, stability, and scene preservation.
Full article255 words · extracted from arxiv.org · click to collapse
Recent video generation is increasingly realistic and controllable, yet video editing remains less developed, particularly for precise local edits that must preserve the original scene dynamics. Video scene text editing replaces text on scene surfaces, such as storefront signs, whiteboards, and product labels, while preserving the surrounding content, motion, and camera dynamics. Although scene text editing is well studied for images, video scene text editing that achieves high visual quality, temporal consistency, and edit locality remains underexplored. Existing resources offer limited paired real-video data, and general video-editing metrics do not directly measure whether the requested text remains correct over time. We introduce ViTeX-Bench, a benchmark suite comprising ViTeX-Dataset and a three-axis evaluation protocol. The dataset contains 387 real-world 720p videos with text-region masks and editing instructions: 230 provide reviewed, pipeline-generated paired edits for training, and 157 form a frozen evaluation split. The protocol evaluates text correctness, visual and temporal quality, and edit locality through 13 metrics, with one primary metric per axis and a Pareto comparison of their trade-offs. OCR calibration, human evaluation, and annotation-sensitivity analyses support the interpretation of these scores. Across eight baselines from four editing families, accurate text, temporal stability, and scene preservation remain difficult to achieve together. We also release ViTeX-Edit-14B, an open-source reference editor fine-tuned on the paired training split with motion-aligned glyph-video conditioning. It achieves CharAcc 0.688, the highest mean among the evaluated video-native editors, and the lowest comparable text-crop Warp among raw editor outputs. ViTeX-Bench provides a reproducible foundation for studying these trade-offs in video scene text editing.
Text extracted automatically; images, tables and formatting may be missing. Original: https://arxiv.org/abs/2609.40356