pre_compile = True时，模型转换错误

模型转换代码

1. from rknn.api import RKNN
   
2. 
   
3. INPUT_SIZE = 64
   
4. 
   
5. if __name__ == '__main__':
   
6. 
   
7.     # Create RKNN object
   
8.     rknn = RKNN()
   
9. 
   
10.     # init runtime environment
    
11.     print('--> Init runtime environment')
    
12.     ret = rknn.init_runtime()
    
13.     if ret != 0:
    
14.         print('Init runtime environment failed')
    
15.         exit(ret)
    
16.     print('done')
    
17. 
    
18.     # Config for Model Input PreProcess
    
19.     rknn.config(channel_mean_value='127.5 127.5 127.5 128', reorder_channel='0 1 2')
    
20.     #rknn.config(channel_mean_value='0 0 0 255', reorder_channel='2 1 0')
    
21. 
    
22.     # Load TensorFlow Model
    
23.     print('--> Loading model')
    
24.     rknn.load_caffe(model='./model/det1.prototxt',
    
25.                       proto='caffe',
    
26.                       blobs='./model/det1.caffemodel')
    
27.     print('done')
    
28. 
    
29.     # Build Model
    
30.     print('--> Building model')
    
31.     rknn.build(do_quantization=False, pre_compile = True, dataset='./dataset.txt')
    
32.     print('done')
    
33. 
    
34.     # Export RKNN Model
    
35.     rknn.export_rknn('./model/det1.rknn')
    
36. 
    
37.     # Release RKNN Context
    
38.     rknn.release()

复制代码错误日志

1. [toybrick@localhost Mtcnn]$ python3 rknn_transfer.py
   
2. /usr/lib64/python3.6/site-packages/h5py/__init__.py:36: FutureWarning: Conversion of the second argument of issubdtype from `float` to `np.floating` is deprecated. In future, it will be treated as `np.float64 == np.dtype(float).type`.
   
3.   from ._conv import register_converters as _register_converters
   
4. --> Init runtime environment
   
5. done
   
6. --> Loading model
   
7. done
   
8. --> Building model
   
9. E Catch exception when building RKNN model!
   
10. T Traceback (most recent call last):
    
11. T   File "rknn/api/rknn_base.py", line 487, in rknn.api.rknn_base.RKNNBase.build
    
12. T   File "rknn/api/rknn_base.py", line 416, in rknn.api.rknn_base.RKNNBase._build
    
13. T   File "rknn/base/ovxconfiggenerator.py", line 146, in rknn.base.ovxconfiggenerator.generate_vx_config_from_files
    
14. T   File "rknn/base/rknnlib/app/helper/striphelper.py", line 101, in rknn.base.rknnlib.app.helper.striphelper.StripHelper.build_network_2
    
15. T   File "rknn/base/rknnlib/app/helper/striphelper.py", line 447, in rknn.base.rknnlib.app.helper.striphelper.StripHelper._pack_data
    
16. T   File "rknn/base/rknnlib/app/helper/vdata_generator.py", line 98, in rknn.base.rknnlib.app.helper.vdata_generator.VDataGenerator.__init__
    
17. T   File "rknn/base/rknnlib/app/helper/vdata_generator.py", line 245, in rknn.base.rknnlib.app.helper.vdata_generator.VDataGenerator._parse_ovx_vdata
    
18. T FileNotFoundError: [Errno 2] No such file or directory: '/tmp/tmp_ta3hlyl/PNet.vdata'
    
19. done
    

复制代码模型文件详见附件

没人帮忙解决一下吗。。。

771488533 发表于 2019-3-8 16:42
没人帮忙解决一下吗。。。

arm64版本的rknn-toolkit暂时还不支持pre_compile，
如果需要pre_compile，建议用x86版本的进行转换。

zhangzj 发表于 2019-3-8 16:59
arm64版本的rknn-toolkit暂时还不支持pre_compile，
如果需要pre_compile，建议用x86版本的进行转换。 ...

mtcnn 的 P-net 模型输入尺寸是变化的，所以模型转换只能够在toybrick 开发板上完成