ZihengWu
ZihengWu
> We expect there to be tiny differences between runs, so the numbers won't be exactly the same. You can try rerunning the training to see what the results are....
findContours return depends on the version of OpenCV using help(cv2.findContours) in python environment to varify the number of parameters returned and rewrite this code "contours, _ = cv2.findContours(src, cv2.RETR_TREE, cv2.CHAIN_APPROX_NONE)"
I have already answered on github,you can check the cv2.findContours return parameter's num and rewrite this '_target_contour = cv3.find_target_contour(_threshold_image)' ------------------ Original ------------------ From: "Dimitrij Klesev";; Send time: Friday, Sep 15,...
额,半年都太忙了,很久没有看自己的代码了。但我思考了一下,如果没有清的话,可能是要一个。
神经网络参数值初始化,这个部分我没有在专栏介绍.参考论文是何凯明的msra初始化方法、当然还有常用的xavier也可以、使用好的初始化能极大加速网络收敛.减少loss直接跑飞的情况.顺便回答一下为什么:论文的出发点是为了保证数据经过层的传输后标准差没有显著的变大或变小、否则bp的过程梯度会反方向变化、从而容易导致爆炸或消失
你看的应该是早期的版本了、在最新版本中、我专门实现了initializer的函数、可供选择、但一般这个问题放在比较后面、实际中网络跑飞了、这个是排查的优先级比较低、而且现在可选的不多、所以我没有专门介绍
tensor/Operator.py里面实现了Conv2D的类
欢迎参考我在知乎写的专栏、完整地介绍了所有的实现和代码细节、希望可以帮助到你 https://zhuanlan.zhihu.com/p/33773140?utm_medium=social&utm_source=wechat_timeline&from=timeline&isappinstalled=0
你理解的没错,但是prediction是一整个batch的结果,prediction[i,label[i]]计算的是第i个数据的Loss,所以这里有一个for循环。注意整个代码中流动的数据第一个维度都是batch
https://www.runoob.com/python/python-func-reduce.html https://docs.python.org/zh-cn/3/library/functools.html 希望可以帮到你 ------------------ Original message ------------------ From: "caoqi95"; Sendtime: Wednesday, Sep 4, 2019 3:32 PM To: "wuziheng/CNN-Numpy"; Cc: "Butterfly "; "Comment"; Subject: Re: [wuziheng/CNN-Numpy] Questions (#1) xxxxx\CNN-Numpy-master\tensor\util.py", line 13,...