tonghe90 / textspotter

324 stars 112 forks source link

@tonghe90 请问训练时,文字识别的ground truth(文字的标签)在哪里输入? #39

Open wenston2006 opened 5 years ago

wenston2006 commented 5 years ago

如题,我在train.pt中没看到文字识别的ground truth,如果没有文字的标签,文字识别部分如何训练呢?

crazysal commented 5 years ago

sample_gt_label_input is one hot vector embedding of char label from groundtruth concatenated to first attention Lstm output check here : for model visualisation http://ethereon.github.io/netscope/#/gist/edc52a3554d5f1e8ee6a24d774de2bd0

Serissa commented 5 years ago

如题,我在train.pt中没看到文字识别的ground truth,如果没有文字的标签,文字识别部分如何训练呢?

你好!请问你有成功训练模型吗?