COCO-Text

Text annotations layered on MS COCO photographs. Large, incidental, and full of attributes (legible or not, printed or handwritten, language).

last verified 2026-09-05

Images63,686 images from MS COCO 2014
AnnotationAxis-aligned word boxes with transcription and attributes (legibility, machine-printed vs handwritten, language); v1 has 173,589 text instances
LanguagesMostly English, with other languages tagged
What it testsScale: text that appears by accident in everyday photos, at every size and quality
How it is scoredICDAR 2015 style IoU at 0.5, computed with the COCO-Text evaluation API; illegible text is don't-care
LicenceAnnotations under the COCO-Text terms; images under the MS COCO / Flickr terms
Downloadhttps://bgshih.github.io/cocotext/
PaperVeit et al., COCO-Text: Dataset and Benchmark for Text Detection and Recognition in Natural Images, 2016

COCO-Text (Veit, Matera, Neumann, Matas and Belongie, 2016) took the MS COCO image set, which was never collected with text in mind, and annotated every piece of text in it. That makes it the closest thing to “text in the wild at internet scale”: most images have no text, some have dozens of tiny words, and a large share is illegible.

What it tests

Recall on incidental text at all scales, and precision in images where nothing should fire. A detector that does well on posed benchmarks can look very different here.

Gotchas

  • Boxes are axis-aligned even for rotated text, so polygon detectors are evaluated against loose ground truth.
  • Version 2 (2017) re-annotated the set; make sure a result names the version.
  • The attribute labels are useful for analysis (how does a model do on handwritten text?) and are often ignored in papers.