-
文档里提到nnie跟qnn支持16bit量化, 实际训练时是否只需要把a_qscheme的bit设置成16即可。
-
你好,我用darknet训练了yolov3后,转换到nnie,用nnie sample,修改类别为两类(人,人头),跑起来后,发现对于大目标person预测正常,但是对于小目标,人头,得分特别低,很多只有0.0?, 我把阈值设置为0.01才把这些框取出来。而在darknet转到caffe后,得分在0.3以上,一切正常。
请问这种现象,是nnie的量化方式导致的吗?使用您的代码,训练后,这种现象会…
-
大佬你好,我在海思上转换了一个yolov4的模型,利用NNIE-LITE进行推理,测试发现识别不准确,如图:
![image](https://user-images.githubusercontent.com/34532224/105473358-00244380-5cd8-11eb-9d88-db36e0c73022.png)
我的anchors配置:
![image](https:…
-
-
我按照你的编译步骤编译,没有报错,但是也没有生成nnie文件
-
想问下这个模型在NNIE上耗时如何呢?
-
大佬好。
请问,你准备写的nnie_yolov5的demo吗?
另外,PART3下的CMakeLists.txt能生成二进制文件吗?
-
NICE work!
I tried to add a new type of backend according to my quantization needs, and I get stuck with some trouble.
I just add my own backend name, say, 'dummy_type' to the class BackendType in p…
-
Hi,
I have a few questions about the Yolov3 and SSD etc quantization.
1-How we can utilize this project for these state of the art methods?
2-Are you planning to add any examples for these n…
-
![image](https://user-images.githubusercontent.com/18262642/118424857-943c6f80-b6fa-11eb-8464-c7d949dca532.png)
首先非常感谢大佬的tutorial,想问下这些函数是不是ssd的前处理和后处理,除了SAMPLE_SVP_NNIE_Forward是调用sdc的算法文件服务,剩余的是自己进行…