Open 6098669 opened 3 years ago
@6098669 Can do you say me what is the exact problem?
因为我跑icdar2015数据集的效果很差,f-score只有78,不知是不是数据集预处理部分没设置好。想请教一下大佬您
能否更新一下icdar2015数据集和yaml文件呢。万分感谢
@921162820 I haven't trained ABCnet on icdar2015 datasets. But, with my experiment, you should config the YAML file and some files suitable with the language of datasets. It's mean that you should add your dataset atAdelaiDet/adet/data/builtin.py
. And you change the size of vocabulary _C.MODEL.BITEXT.VOC_SIZE =
at line 101 of AdelaiDet/adet/data/builtin.py
. You can read more at original abcnet. And, you also should refer to that issue if you want to train with the custom dataset.
I think f1-score 78 is good. I'm glad to help you. Ps: You can write in English. I don't know the Chinese language. I'm sorry
。。