Performance on TextOCR Dataset

[jkcg-learning]

Motivation

Improve the benchmark performance of all algorithms based on TextOCR dataset released by Facebook AI research team

Related resources https://textvqa.org/textocr

Overview TextOCR requires models to perform text-recognition on arbitrary shaped scene-text present on natural images. TextOCR provides ~1M high quality word annotations on TextVQA images allowing application of end-to-end reasoning on downstream tasks such as visual question answering or image captioning.

Statistics 28,134 natural images from TextVQA 903,069 annotated scene-text words 32 words per image on average

[cuhk-hbsun]

Thanks for your suggestion. And we will take it into our July plan.

[jkcg-learning]

Team, is this in consideration for the next release ?

[gaotongxiao]

We already support TextOCR dataset now (https://mmocr.readthedocs.io/en/latest/datasets.html)

[jkcg-learning]

Thanks for adding this dataset for the purpose of training...

Shall we also expect a model checkpoint particularly trained based on this dateset from the team..

[gaotongxiao]

Currently we only have DBNet pretrained on TextOCR. Do you have any requests for the model type and the specific datasets that it is pretrained on? We may add that to our plan if we believe that it also benefits our community.

[jkcg-learning]

https://mmocr.readthedocs.io/en/latest/textdet_models.html#icdar2015

Is it possible to update the DBNet model zoo with the details of your model training and the metric levels for TextOCR dataset ..