error to convert ssd_mobilenet_v2_coco_2018_03_29

[复制链接] · 发表于 2019-2-13 12:25:28

I modifed ssd-model.py to conver convert ssd_mobilenet_v2_coco_2018_03_29 to RKNN as below.

rknn.load_tensorflow(tf_pb='./ssd_mobilenet_v2_coco_2018_03_29/frozen_inference_graph.pb')
inputs=['FeatureExtractor/MobilenetV2/MobilenetV2/Conv2d_0/BatchNorm/batchnorm/mul_1'],

But I got the following errors:

--> Loading model
Catch exception when loading tensorflow model: ./frozen.pb!
Traceback (most recent call last):
  File "/usr/local/lib/python3.5/dist-packages/rknn/api/rknn.py", line 50, in load_tensorflow
std_values=std_values)
  File "rknn/api/redirect_stdout.py", line 67, in rknn.api.redirect_stdout.redirect_stdout.func_wrapper
  File "rknn/api/redirect_stdout.py", line 68, in rknn.api.redirect_stdout.redirect_stdout.func_wrapper
  File "rknn/api/rknn_base.py", line 138, in rknn.api.rknn_base.RKNNBase.load_tensorflow
  File "rknn/base/rknnlib/converter/convert_tf.py", line 93, in rknn.base.rknnlib.converter.convert_tf.convert_tf.__init__
  File "rknn/base/rknnlib/converter/tensorflowloader.py", line 53, in rknn.base.rknnlib.converter.tensorflowloader.TF_Graph_Preprocess.__init__
AttributeError: 'NoneType' object has no attribute 'op'

done
--> Building model
Catch exception when building RKNN model!
Traceback (most recent call last):
  File "/usr/local/lib/python3.5/dist-packages/rknn/api/rknn.py", line 159, in build
ret = self.rknn_base.build(do_quantization=do_quantization, dataset=dataset, pack_vdata=pre_compile)
  File "rknn/api/redirect_stdout.py", line 67, in rknn.api.redirect_stdout.redirect_stdout.func_wrapper
  File "rknn/api/redirect_stdout.py", line 68, in rknn.api.redirect_stdout.redirect_stdout.func_wrapper
  File "rknn/api/rknn_base.py", line 404, in rknn.api.rknn_base.RKNNBase.build
  File "rknn/api/rknn_base.py", line 703, in rknn.api.rknn_base.RKNNBase._quantize
  File "rknn/base/tensorzonex.py", line 213, in rknn.base.tensorzonex.get_data_file
  File "/usr/lib/python3.5/posixpath.py", line 122, in splitext
return genericpath._splitext(p, sep, None, extsep)
  File "/usr/lib/python3.5/genericpath.py", line 118, in _splitext
sepIndex = p.rfind(sep)
AttributeError: 'NoneType' object has no attribute 'rfind'

done
Export RKNN model failed!
done

Any idea why this happens?

Pangu,

只看该作者 · 发表于 2019-2-13 15:52:35

Please use rknntoolkit 0.9.8 try again, initializing with verbose argument, and it will print more error info to us.

只看该作者 · 发表于 2019-2-14 01:23:19

Below is output of rknn-toolkit-v0.9.8 with verbose=True:

D import clients finished
E Catch exception when loading tensorflow model: ./ssd_mobilenet_v2_coco_2018_03_29/frozen_inference_graph.pb!
T Traceback (most recent call last):
T File "rknn/api/rknn_base.py", line 143, in rknn.api.rknn_base.RKNNBase.load_tensorflow
T File "rknn/base/rknnlib/converter/convert_tf.py", line 93, in rknn.base.rknnlib.converter.convert_tf.convert_tf.__init__
T File "rknn/base/rknnlib/converter/tensorflowloader.py", line 53, in rknn.base.rknnlib.converter.tensorflowloader.TF_Graph_Preprocess.__init__
T AttributeError: 'NoneType' object has no attribute 'op'
2019-02-13 08:57:06.514403: I tensorflow/core/platform/cpu_feature_guard.cc:141] Your CPU supports instructions that this TensorFlow binary was not compiled
to use: AVX2 FMA
E Catch exception when building RKNN model!
T Traceback (most recent call last):
T File "rknn/api/rknn_base.py", line 470, in rknn.api.rknn_base.RKNNBase.build
T File "rknn/api/rknn_base.py", line 874, in rknn.api.rknn_base.RKNNBase._quantize
T File "rknn/base/tensorzonex.py", line 213, in rknn.base.tensorzonex.get_data_file
T File "/usr/lib/python3.5/posixpath.py", line 122, in splitext
T return genericpath._splitext(p, sep, None, extsep)
T File "/usr/lib/python3.5/genericpath.py", line 118, in _splitext
T sepIndex = p.rfind(sep)
T AttributeError: 'NoneType' object has no attribute 'rfind'
E RKNN model is None

只看该作者 · 发表于 2019-2-14 17:03:06

Are you able to find out why ssd_mobilenet_v2_coco_2018_03_29 model can not work with RKNN tool?

Thanks,

只看该作者 · 发表于 2019-2-15 10:27:32

This model, http://download.tensorflow.org/m ... o_2018_01_28.tar.gz is OK.
But this one, v2, http://download.tensorflow.org/m ... o_2018_03_29.tar.gz is not OK.
The error is below using RKNN-0.9.8:

D save dump info to: ./test.log
--> Loading model
D import clients finished
E Catch exception when loading tensorflow model: ./ssd_mobilenet_v2_coco_2018_03_29/frozen_inference_graph.pb!
T Traceback (most recent call last):
T File "rknn/api/rknn_base.py", line 143, in rknn.api.rknn_base.RKNNBase.load_tensorflow
T File "rknn/base/rknnlib/converter/convert_tf.py", line 93, in rknn.base.rknnlib.converter.convert_tf.convert_tf.__init__
T File "rknn/base/rknnlib/converter/tensorflowloader.py", line 53, in rknn.base.rknnlib.converter.tensorflowloader.TF_Graph_Preprocess.__init__
T AttributeError: 'NoneType' object has no attribute 'op'
done
--> Building model
2019-02-14 18:05:06.858750: I tensorflow/core/platform/cpu_feature_guard.cc:141] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2 FMA
E Catch exception when building RKNN model!
T Traceback (most recent call last):
T File "rknn/api/rknn_base.py", line 470, in rknn.api.rknn_base.RKNNBase.build
T File "rknn/api/rknn_base.py", line 874, in rknn.api.rknn_base.RKNNBase._quantize
T File "rknn/base/tensorzonex.py", line 213, in rknn.base.tensorzonex.get_data_file
T File "/usr/lib/python3.5/posixpath.py", line 122, in splitext
T    return genericpath._splitext(p, sep, None, extsep)
T File "/usr/lib/python3.5/genericpath.py", line 118, in _splitext
T    sepIndex = p.rfind(sep)
T AttributeError: 'NoneType' object has no attribute 'rfind'
done
--> Init runtime environment
done
--> Running model
E RKNN model is None
done
inference result:  None
Traceback (most recent call last):
  File "./ssd2.py", line 90, in <module>
predictions = outputs[1].reshape((1, NUM_RESULTS, 4))
TypeError: 'NoneType' object is not subscriptable

Can you download http://download.tensorflow.org/m ... o_2018_03_29.tar.gz to reproduce it?

Thanks,

只看该作者 · 发表于 2019-2-15 13:13:38

pangu 发表于 2019-2-15 10:27
This model, http://download.tensorflow.org/models/object_detection/ssd_mobilenet_v1_coco_2018_01_28. ...

我今天在测试时也发现了，好像只有 V1 版本能转， V2 或者 V1 其它数据集的都又这个问题，是不支持某些 OP 么？

只看该作者 · 发表于 2019-2-16 10:04:35

Agreed. I even try export_inference_graph on ssd_mob ile_v1_coco below and use the frozen pb with ssd.py, I got the same error again.

python3 ./src/models/research/object_detection/export_inference_graph.py \
--input_type image_tensor \
--pipeline_config_path ssd_mobilenet_v1_coco_2018_01_28/pipeline.config \
--trained_checkpoint_prefix ssd_mobilenet_v1_coco_2018_01_28/model.ckpt \
--output_directory object_detection_graph

It seems rknn model converter tool is not mature and hope RK eng can fix it asap. Otherwise, we can not use it for our own trained model

只看该作者 · 发表于 2019-2-18 11:19:22

这个问题解决了么？

只看该作者 · 发表于 2019-2-18 12:02:54

mobilenet v2的input和v1是不一样的，请用参考下面代码

    

rknn.load_tensorflow(tf_pb='./ssd_mobilenet_v2_coco_2018_03_29/frozen_inference_graph.pb',

                         inputs=['FeatureExtractor/MobilenetV2/MobilenetV2/input'],

                         outputs=['concat', 'concat_1'],

                         input_size_list=[[300, 300, 3]])

复制代码