search

Yuliang-Liu / Curve-Text-Detector

This repository provides train＆test code, dataset, det.&rec. annotation, evaluation script, annotation tool, and ranking.
639 stars 155 forks source link

label style is different in training and test data #48

Closed neptune4year closed 4 years ago

neptune4year commented 4 years ago

I find that the label in the training set is usually one polygon for one word, but that in the test set is one polygon for one whole line. Is that true? Do you have correct labels? Please refer to image 0468 in training set and image 1233 in test set.

zhangshuaitao commented 4 years ago

@neptune4year, all annotations are based on text lines.The question you said should be a special case. We will try to ensure that all annotations are based on text lines. Thanks for your attention.