search

opconty / Transformer_STR

PyTorch implementation of my new method for Scene Text Recognition (STR) based on Transformer,Equipped with Transformer, this method outperforms the best model of the aforementioned deep-text-recognition-benchmark by 7.6% on CUTE80.
https://mp.weixin.qq.com/s/a_ahIwxiCaO7Bxmj81HUTw
224 stars 53 forks source link

The Recognition Accuracy on other datasets? #3

Open zobeirraisi opened 4 years ago

zobeirraisi commented 4 years ago

What is the performance on other datasets, and why only CUTE80 has been improved?

opconty commented 4 years ago

Because we mainly focus on curved text recognition, and we think this is the most challenging task, and we also refine the model on this aspect.

On Sun, Jun 21, 2020 at 9:11 AM Zobeir Raisi notifications@github.com wrote:

What is the performance on other datasets, and why only CUTE80 has been improved?

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/opconty/Transformer_STR/issues/3, or unsubscribe https://github.com/notifications/unsubscribe-auth/ACMUAWTCFPXVCSU6GANKTP3RXVM4LANCNFSM4ODT7HHQ .

zobeirraisi commented 4 years ago

Thanks for your reply, it means that after training on synthetic dataset the model again fine-tuned on CUTE-80 or real-world dataset?

zobeirraisi commented 4 years ago

How many epochs have been used for training?

opconty commented 4 years ago

Hi:

I reckon about 3 epochs for training. Do not remember the exact number. We do not fine-tune on real-world dataset, just pick the most skewed dataset as valid-set. and you can train this model from scratch to figure out the result.

Best.