-
@argman @zxytim I try to train the model by icdar2015 Incidental Scene Text data set for Text Localization (1000 image samples).
_The training script is as follows:_
**python multigpu_train…
-
## 실험1
* 개요
ICDAR17_train으로 학습된 모델(LBscore : 0.6749)에 polygon으로 annotation된 데이터(ART)를 추가로 fine-tunning시켜 주었다.
| | LB public | LB private |data(pre-trained)|data(fine-tunning)|additional epoch|
…
-
Hi,
Please can you tell the steps taken for pre-procesing synthtext labels ??
your model uses fixed max length of 25 but synthtext dataset has boxes with labels length(number of characters per…
-
Could you relaase the models that you have trained on icdar15 or icdar 17?
Thank you very much.
-
I think the following data type should be uint64 not uint8.
https://github.com/argman/EAST/blob/dca414de39a3a4915a019c9a02c1832a31cdd0ca/icdar.py#L465
because https://github.com/argman/EAST/blob/dc…
-
I am using the pre-trained model for testing on my own images. I had downloaded model_IC15_89.pth using the link https://drive.google.com/file/d/1nv-ZjbYBj8ePZRa_fAhbHvzm7HqSxPWK/view?usp=sharing and …
-
具体情况如下:
loss: nan (nan) loss_classifier: 0.2950 (1.0475) loss_box_reg: 0.0002 (0.0087) loss_objectness: nan (nan) loss_rpn_box_reg: nan (nan)
求各位大佬指点迷津,感激不尽!
@mjq11302010044
下面是我使用的bbox格式…
-
Hi,
Thanks for your implementation of the EAST paper. I am trying to reimplement EAST in PyTorch for some work. Could you please explain the process you have followed to recover the rectangle corne…
-
"Our model was trained on 3,000 natural images, including 229 images
from the ICDAR 2013 training set. We collected the other images ourselves and
manually labelled them with text line bounding boxe…
-
How to create ICDAR-13 data for training. I am getting the following error :
![error](https://user-images.githubusercontent.com/20046403/27384425-6cf6a628-56ac-11e7-9853-51a74267b4dc.png)