The ICDAR2015-TextSR dataset
This dataset was originally presented for the ICDAR2015 Competition on Text Image Super-Resolution. The released version contains supplementary materials (original images, annotations).
Download
To get access to the dataset, you first need to register by sending an e-mail to icdar-sr2015 (at) liris.cnrs.fr with the following informations :
- *Name of the participant(s) / team :
- *University / Organization :
- Preferred contact e-mail address :
- Motivation/Interest :
We will send you a download link as soon as we receive your informations.
Description
The released dataset contains:
- a training set: 567 HD (original images), LR (/4) & HR (/2) images with their transcription (~12500 characters)
- a test set: 141 HD, LR & HR images with their transcription (2929 characters)
- An informative file and some useful tools
Copyright
All the images and transcriptions contained in the ICDAR2015-SR-dataset constitute a database which is the property of Orange (© 2013-2015), and are licensed under the Open Database Licence (ODbL). By registering you agree to abide by all terms of the ODbL license applied to the ICDAR2015-SR-dataset.
Samples from the training dataset
Here are some samples from the training dataset : the LR image, its upscaled version using bicubic interpolation, the HR image and the manual transcription. To get access to the complete dataset, you have to proceed to registration.
| ↓ Ground Truth ↓ | |||
|---|---|---|---|
| LR | Bicubic | HR | Transcription |
|
|
|
Musique_:_"Celebration"_(_EMI_MUSIC_)_/_R.Bell_R.Bell,_G.Brown
|
|
|
|
Le_piratage_nuit_a_la_creation_artistique
|
|
|
|
DIE_ZAUBERFLOTE
|
|
|
|
Budget_350E
|
|
|
|
Adherez_vite_sur
|
|
|
|
Objectif_HEC
|
|
|
|
"Can't_Remember_to_Forget_You"
|
