-
First,thanks for the author's ditribution for this project ! Can anyone tell me the "demo.py" for this porject, I can only get test mode for test datasets.
-
In the paper, only two contrasting decoding methods were shown: AR (one character at a time) and non-AR (all characters at once). But in theory, PARSeq supports semi-autoregressive decoding and anythi…
-
When I try to pre-train the vision model and ABINet I found that there are 3 evaluation datasets that contain corrupted image. Any fix for this ? Or do I just remove it from the configs ?
Below ar…
-
In order to improve the accuracy of scene text recognition, some existing methods first rectify the text image(e.g. Iterative Image Rectification) and then recognize text. ABINet does not adopt the m…
-
Hi!
I’m very impressed that ABINet’s ability to recognize words. I am starting to use it to help identify words in old newspapers when Tesseract cannot accurately identify a word and returns results…
-
I have Anaconda installed on my windows machine, so I only followed the instructions below `Activate conda environment`. So create the new environment, activate it and install the dependencies via cop…
-
As mentioned in the paper, you use 20% training data(around 16M*0.2 = 3.2M) to train the model. I have some questions about it.
Previously, the baseline model ABINet consists of three stages: vision …
-
你好,请问如果要增加训练epoch次数,应该怎么操作?我试着修改了train_abinet.yaml中的epoch数,训练时能够正常显示更新后的epoch数,但是每次训练仍然到epoch=10时自动结束进程,也未有任何错误提示,这是怎么回事呢?
-
想查看整个ABINet的流程,发现训练过程中很多细节都被fastai封装了,请问要如何查看这些细节呢?
例如loss的forward,不太明白loss是什么时候调用的,然后loss的参数有哪些,还有训练期间网络模型的输入和输出也是不可见的。
-
I have trained the model 4 epochs, but I find the learning rate inappropriate. Thus I change the learning rate schedule, and then I retrain the network by resuming from the model with trained 3 epoch…