Shual

Results 19 comments of Shual

@backtime92 I also care about this issue. My wechat number is ShuYang097. Thank you!

@rremani Thanks a lot!

SynthText only has '.mat' for ground truth labeling. How can I know the structure of file '.txt' for gt labeling?

> 我直接在2015上训练的,剪枝使模型大小缩小一半,再finetune,hmean都可以达到82多,不过我是在v100上训练的 ------------------------------ 请问您的batchsize是多大。我想有可能是batchnorm的关系,我是batch10在双卡上训练的。

@BowieHsu @JiasiWang 我用了SynthText 40g做的tf文件,预训练90000轮以后,因为finetune_ic15.json里面"finetune_model": "../exp/sgd/checkpoint"(默认)跑不通,我改成了"finetune_model": "../exp/sgd/checkpoint-90000"。接下来训练10000轮以后。在ic15测试集上面跑出的结果只有 Recall | Precision | Hmean 59.56 % | 63.47 % | 61.45 % 为什么没有达到75%呢? 道友盼回复,感谢大佬!

改成batch-size32 依然hmean,61%左右。

我拿预训练模型跑测试,不经过finetune,结果是hmean49%

https://github.com/yulequan/PU-Net/issues/5 maybe we should recompile the .so file or re-install tensorflow