nnom icon indicating copy to clipboard operation
nnom copied to clipboard

A higher-level Neural Network library for microcontrollers.

Results 88 nnom issues
Sort by recently updated
recently updated
newest added

想请教下,在模型转h文件时,必须要加入x_test做转换么? 如果我采用训练完备的h5模型,是无法直接转换使用。 关于直接采用训练完备的模型,如何快速转weight.h,而不需要重新训练。

Hi, It seems that CMSIS-NN has new versions like [v3.1+](https://arm-software.github.io/CMSIS_5/NN/html/index.html). Is there any chance for NNoM to support newer CMSIS-NN? Regards, yf13

使用relu6这类带截断的激活函数训练模型,量化后输出结果出错,想请教一下是什么原因?

目前量化脚本好像只支持DepthwiseConv2D,缺少DepthwiseConv1D

各位大佬有在移植过程中,有遇到了内存泄露的问题嘛,定位出来在nnom的output层函数

Hi I am confused about this code for GRU MAC calculation ![image](https://user-images.githubusercontent.com/41681967/175035615-0a0693e5-7740-4ba3-a920-55426c33be49.png) What I know about LSTM MAC calculation is realized as following picture: ![image](https://user-images.githubusercontent.com/41681967/175036140-a3242b82-cd00-4b0e-b213-033c45bd092c.png) Can you give more information?...

NNOM 支援深度可分离卷积嗎?或是可以透過depthwise(DW)和pointwise(PW)两个部分结合起来嗎? 謝謝

RNN量化损失相对较大,可以增加支持16bit算子吗?

你好 使用CMSIS-NN 後端進行加速 我是在rt-thread 上進行 直接使用上面的CMSIS Package ![image](https://user-images.githubusercontent.com/76082025/166668779-303b02ff-9c2e-4626-9b90-d6c0ce7691af.png) 有添加兩個 ' ARM_MATH_CM7, __FPU_PRESENT=1' 因為我是使用F746(不知道是不是使用ARM_MATH_CM7) ![image](https://user-images.githubusercontent.com/76082025/166669559-6517e9de-5f5c-4d42-93ed-f932a985f34f.png) 如果我有任何錯的地方請麻煩告訴我 非常感謝

Hello, I would like to do Model optimization to trim non-essential weights at model , this not supported by the pruning API provided by tensorflow , is there a trimming...