Chow Yue
Chow Yue
这是文字序列识别，模型的最后是通过将特征图的每一列分别展开输入全连接层识别的，加入你输入的图像最终得到的特征图有70列就会有70个识别结果，然后通过CTC合并得到最终结果，所以和你说的50%的理论不相符
模型预测出的序列结果并不是一个整体，而是一个由70列预测结果组成的序列（使用CTC之前），你训练的两张图里面其实可能包含着70*2个不同的类别，自然不是随机猜一个50%
> @liguangcheng Got it, thanks. @liangjianwen I still got the same problem with 2.8.1 even though I met the two conditions mentioned by @liguangcheng , does it work for you?
i agree with @Mte90 ,that would be a useful one > Ideally is enough to have a function that print `3 TODO | 1 FIXME` so it can be used...
I also found missing entities for my Pawmate Pet Feeder, 1- Verify that you followed the following guides: a) i configured Tuya OpenAPI credentials: d) i have enable all dpcodes...
Hi， Thank you for your response. While the `meal_plan` field is certainly something I need, I'm more concerned about why the `status` field isn't appearing on my device's dashboard. I...
Excellent! I have obtained all the entities I need except for meal_plan. This is tremendously good news for me. Thank you so much!