CenterNet精度量化效果很差

jefferyzhang 发表于 2020-1-2 12:01
从图是看不出精度问题的，只能推断出很可能是精度问题。。。
检不捡的出还得看你NMS阈值设置多少 ...

版主大大，再请教两个问题：
1. 如何进行通道级量化？是将cfg文件中各项method参数由layer改成channel吗？
2. config file的作用是什么？

1. ...
   
2. I Config File "/home/lsc/anaconda3/envs/RKNN/lib/python3.6/site-packages/rknn/base/RK1808_PID0X82" load/generated successfully
   
3. I Initialzing network optimizer by /home/lsc/anaconda3/envs/RKNN/lib/python3.6/site-packages/rknn/base/RK1808_PID0X82 ...
   
4. ...
   

复制代码

per_channel的量化我们不支持

混合量化，将一些层从cfg文件中去掉，方法改成delete，生成模型之后运行会报这个错误。

1. 
   
2. Create RKNN model fail!
   
3. E Catch exception when init runtime!
   
4. E Traceback (most recent call last):
   
5. E   File "rknn/api/rknn_base.py", line 788, in rknn.api.rknn_base.RKNNBase.init_runtime
   
6. E   File "rknn/api/rknn_runtime.py", line 303, in rknn.api.rknn_runtime.RKNNRuntime.build_graph
   
7. E Exception: RKNN init failed. error code: RKNN_ERR_DEVICE_UNAVAILABLE
   
8. W We need do some initial work, it will increase call time of inference for the first time.
   
9. Create RKNN model fail!
   
10. E Catch exception when init model!
    
11. E Traceback (most recent call last):
    
12. E   File "rknn/api/rknn_base.py", line 901, in rknn.api.rknn_base.RKNNBase.inference
    
13. E   File "rknn/api/rknn_runtime.py", line 303, in rknn.api.rknn_runtime.RKNNRuntime.build_graph
    
14. E Exception: RKNN init failed. error code: RKNN_ERR_DEVICE_UNAVAILABLE
    
15. Traceback (most recent call last):
    
16.   File "ctdet.py", line 10, in <module>
    
17.     det.run(img)
    
18.   File "/home/lsc/rknn-toolkit-v1.3.0/centernet/ctdet_rknn_v2/detector.py", line 93, in run
    
19.     output, dets = self.process(images)
    
20.   File "/home/lsc/rknn-toolkit-v1.3.0/centernet/ctdet_rknn_v2/detector.py", line 44, in process
    
21.     hm = output[3].reshape((1,80,128,128))
    
22. TypeError: 'NoneType' object is not subscriptable

复制代码