search

cs-chan / Total-Text-Dataset

Total Text Dataset. It consists of 1555 images with more than 3 different text orientations: Horizontal, Multi-Oriented, and Curved, one of a kind.
BSD 3-Clause "New" or "Revised" License
747 stars 140 forks source link

Can you offer me your newest test set ground truth? #27

Closed gw00295652 closed 4 years ago

gw00295652 commented 4 years ago

I can't find the newest test data set ground truth. this is my email : taoju825538334@163.com thanks!

WeihongM commented 4 years ago

Is the ground-truth of test data renewed also? Or still the same as the legacy version? @chunchet-ng @cs-chan

ckchng commented 4 years ago

Hi guys, thanks for the interest! We do not release a new version of the test set ground truth because 1) there is no need of standardising the length of the ground truth vertices for testing purpose, it was proposed to facilitate training only, 2) a new version of ground truth would make the previous benchmarks irrelevant.

So if your intention is testing/benchmark only, there is no need for the new ground truth. Do let me know if you intend to use it for other unforeseen purposes, and we shall discuss further. Cheers!

WeihongM commented 4 years ago

@ckchng Thanks for your reply. I am not familiar with this dataset. Can you share the main difference when we use fixed 10 vertices to represent the polygon compared with former ground-truth?

ckchng commented 4 years ago

The number of vertices in the former ground-truth is unfixed. It could be a problem if you want to train a regression-based deep nets. The new ground-truth is annotated with a guided mechanism to remove biases too.

It's all described in the following paper https://link.springer.com/article/10.1007/s10032-019-00334-z

WeihongM commented 4 years ago

Got it, thanks for your reply.