我打算用Tengine sdk转换量化yolov3-tiny,但是在使用convert tool进行转换的时候程序出错(PC运行)：
使用的weight是
https://pjreddie.com/media/files/yolov3-tiny.weights
cfg是：
请教下这个出错的原因是什么呢？
使用的weight是
https://pjreddie.com/media/files/yolov3-tiny.weights
cfg是：
工具应该是没问题的，yolov3我测过的，可以正常运行的。
我只是换成yolov3-tiny出错了
不过我有看到convert tool的文档是支持yolov3-tiny的
但是Tengine另一份spec写的NPU支持网络没有yolov3-tiny,不过这里不支持结合前面看起来是不支持后面的量化或者什么？
@machengsen 我这边测试是可以的
$ ./convert_tool -f darknet -m ~/Downloads/yolov3-tiny.weights -p ~/yan/tmp/yolov3-tiny.cfg -o yolov3-tiny.tmfile
---- Tengine Convert Tool ----
Version : v1.0, 15:43:59 Jun 24 2021
Status : float32
major: 0, minor: 2, revision: 0, seen: 0, transpose: 0
Create tengine model file done: yolov3-tiny.tmfile
用的是你的cfg和weights
@Frank I have the same problem when converting my custom Yolov3 model using Tengine. I have just modifid the config file activation function with relu instead of leaky.
Please let me know the solution.
One month before, it had worked for me. Now it is not working.
@Akkisony I think you can try the model in our docs. If you work, and you model is not, maybe it include some unsupport layers
@Frank As I said, I just modified the activation function and nothing much.
Isn’t relu activation function supported?
@Frank Where can I find the
The model with relu activation function was converted using Python API - NPU SDK 
. However, it was not possible with Tengine SDK. 
Just wanted to inform you about it.
@Akkisony OK, thanks. Python API uses amlogic’s SDK, tengine uses its own SDK, both of which define their own supported layers, and the layers supported by the two overlap, which is normal.
