ClapperText: A Benchmark for Text Recognition in Low-Resource Archival Documents

TL;DR

ClapperText addresses OCR for low-resource archival documents by providing a word-level, rotated-rectangle annotated benchmark derived from WWII-era clapperboards. It features 127 video segments, 9,813 annotated frames, 94,573 word instances, with 67% handwritten and 1,566 occluded, plus cropped word images and semantic labels. The authors benchmark six text-recognition and seven text-detection models under zero-shot and fine-tuned regimes, demonstrating a clear domain gap and meaningful gains from few-shot adaptation. This resource enables robust OCR research for degraded archival media and motivates future domain adaptation and temporal-context modeling across frames.

Abstract

This paper presents ClapperText, a benchmark dataset for handwritten and printed text recognition in visually degraded and low-resource settings. The dataset is derived from 127 World War II-era archival video segments containing clapperboards that record structured production metadata such as date, location, and camera-operator identity. ClapperText includes 9,813 annotated frames and 94,573 word-level text instances, 67% of which are handwritten and 1,566 are partially occluded. Each instance includes transcription, semantic category, text type, and occlusion status, with annotations available as rotated bounding boxes represented as 4-point polygons to support spatially precise OCR applications. Recognizing clapperboard text poses significant challenges, including motion blur, handwriting variation, exposure fluctuations, and cluttered backgrounds, mirroring broader challenges in historical document analysis where structured content appears in degraded, non-standard forms. We provide both full-frame annotations and cropped word images to support downstream tasks. Using a consistent per-video evaluation protocol, we benchmark six representative recognition and seven detection models under zero-shot and fine-tuned conditions. Despite the small training set (18 videos), fine-tuning leads to substantial performance gains, highlighting ClapperText's suitability for few-shot learning scenarios. The dataset offers a realistic and culturally grounded resource for advancing robust OCR and document understanding in low-resource archival contexts. The dataset and evaluation code are available at https://github.com/linty5/ClapperText.

Figures (9)

• Figure 1: Visual examples of annotated keyframes in ClapperText. Each subfigure shows eight frames, illustrating variation in layout, handwriting style, and image quality across splits.
• Figure 2: Distribution of frame counts per video segment in ClapperText.
• Figure 3: Keyframe samples from five ClapperText videos. Each column illustrates a distinct intra-video variation: (1) minimal change across dense frames, (2) background variation outside the clapperboard, (3) color shift and damage obscuring text, (4) clapperboard motion affecting visibility, and (5) brightness fluctuation.
• Figure 4: Annotation pipeline: historians transcribe shot-level text; computer vision annotators label text instances in video frames.
• Figure 5: Examples of annotated data in ClapperText. (a) shows word-level annotations with semantic categories overlaid on a keyframe. (b) presents cropped word images extracted from those annotations, including handwritten (top) and printed (bottom) examples.