Closed seekingdeep closed 3 years ago
The paper shows promising results of 87.3% F-measure on icdar2015, while your implementation scores ~ 83.5% How to achieve similar results to the main implementation? what is the difference between your implementation and the main?
In order to improve the effect, the author used Deformable convolution in the backbone and pre-trained on the sythtext data set, but it did not do in this project
ok
The paper shows promising results of 87.3% F-measure on icdar2015, while your implementation scores ~ 83.5% How to achieve similar results to the main implementation? what is the difference between your implementation and the main?