add gen_image, img2lmdb

[ZhangXinNan]

• add gen_image.py generate image for training
• create_dataset.py is from crnn
• create_dataset_main.py image 2 lmdb

[meijieru]

Training examples mostly come from standard datasets, and there are various fantastic text image synthesize engine. So I don't think we should have a image generator here.

Also, images -> lmdb comes from origin repository, there is no need for a duplicate.

Thanks for your PR.

[jxlijunhao]

@ZhangXinNan while have you test the generated chinese character could be save to lmdb, currently some of us have faced the "TypeError: Won't implicitly convert Unicode to bytes; use .encode()" errors.

[daoqinzi]

@jxlijunhao 我也遇到这个问题了，发现是python版本的原因，改成python2.7就好了

[gittigxuy]

@daoqinzi 可以把自己的数据集，含有多个中文的区域的图片制作成lmdb格式的文件吗？如何做呢？可否详细说明一下

[daoqinzi]

@gittigxuy 可以先按照 @ZhangXinNan 的 gen_image create_dataset_main 跟一边代码就清楚些了