Shangbang Long

Some errors in the 800K datasets: oversized word/char box, missing labels

8

Hi, I am having similar problems as those discussed in #13 #15. I am using the pre-generated 800K dataset to train a model, and found that there exist the following...

(Solved) Found one bug in 4.22 and fixed

3

- Operating System: Ubuntu 16.04 - UE4 Version: 4.22 - UnrealCV Version: diva-branch - Client (python2, 3 or matlab): N/A - Problem Description: While compiling a clean diva-branch that was...

Thanks for the great work! My question is as titled.

How to reproduce results on IC15?

1

IC15 evaluation requires the use of vocabulary. How do you make use of that?