oranusa
oranusa
> There is no problem when 1000 examples are provided . > but when i use 5000 examples to train there is a error: > > Traceback (most recent call...
> 我試過albert+bilstm+crf及單純albert做序列預測,我發現單純的albert有比較好的效果,或許你可以嘗試使用單純的albert� 我是直接，albert+crf，参数比+bilstm少了很多，效果也显著。不知道你说的单纯albert，最后分类其接的是什么，还请给点指点。
and in **_face_detector.py_** code, the function **_precess_** align face_image(it doesn't seem to), but I test some image, get this result:  
have u ever try to another way to indetify face, for example: after get the 512-D feature, then compute the distance withe face database. compare with a classifer, which way...
我有遇到过类似的问题，我是自己的数据，一开始训练ap是有上升的，后面数据做了一些改动，再训练就不会动了。然后换其他的模型，和数据重新跑一次，再切回来就好了，不知道什么问题，可能是中间过程有什么cache影响了吧？
> http://www.blog.zhxing.online/#/readBlog/372 最新的文章，文本摘要的。 强👍！！。max_len=256，但是数据集基本都是很长的，这样直接截断，应该对文本输出的内容影响很大吧？
> > > > 这种只能人工预先处理文档，切分文档中段落为qa对，然后补充上下文了吧。 请问，切分为qa对之后，想知道只是出处（原始文档位置），那由该如何处理呢？
> @1093842024 I test the script on CPU(1 core), the inference time is about 70ms. did u test the time of prior_box part, it's need more time than net forward.
> When it gets converted to a tensor, the default type is float and the default action in Python will normalise it. However, it won't do this automatically in C++....
> That's the mean being subtracted from the image as an unsigned char array, prior to normalisation. > That BGR value becomes the zero of the normalised values. > >...