很简单的模型rknn.load_tensorflow出错，该怎么升级

[复制链接] · 发表于 2020-7-27 17:53:44

D:\rknn-toolkit-v1.3.2\examples\tensorflow\ts> cmd /C "C:\Users\liufeng\Anaconda3\envs\RKNN_TOOL\python.exe c:\Users\liufeng\.vscode\extensions\ms-python.python-2020.7.96456\pythonFiles\lib\python\debugpy\launcher 8665
-- d:\rknn-toolkit-v1.3.2\examples\tensorflow\ts\ts.py "
--> Loading model
C:\Users\liufeng\Anaconda3\envs\RKNN_TOOL\lib\site-packages\onnx_tf\common\__init__.py:87: UserWarning: FrontendHandler.get_outputs_names is deprecated. It will be removed in future release.. Use node.outputs instead.
  warnings.warn(message)
'python3' 不是内部或外部命令，也不是可运行的程序
或批处理文件。
E Try match Add add failed, catch exception!
E Catch exception when loading tensorflow model: ./ts.pb!
E Traceback (most recent call last):
E File "rknn\base\RKNNlib\converter\convert_tf.py", line 616, in rknn.base.RKNNlib.converter.convert_tf.convert_tf.match_paragraph_and_param
E File "rknn\base\RKNNlib\converter\convert_tf.py", line 380, in rknn.base.RKNNlib.converter.convert_tf.convert_tf._tf_try_match_ruler
E File "<string>", line 1, in <module>
E File "rknn\base\RKNNlib\converter\convert_tf.py", line 134, in rknn.base.RKNNlib.converter.convert_tf.convert_tf.shape_pick
E File "rknn\base\RKNNlib\converter\tensorflowloader.py", line 205, in rknn.base.RKNNlib.converter.tensorflowloader.TF_Graph_Preprocess.query_tensors
E File "rknn\base\RKNNlib\converter\tf_util.py", line 371, in rknn.base.RKNNlib.converter.tf_util.TFProto_Util.query_tensor
E File "rknn\base\RKNNlib\converter\tf_util.py", line 374, in rknn.base.RKNNlib.converter.tf_util.TFProto_Util.query_tensor
E File "rknn\base\RKNNlib\converter\tf_util.py", line 407, in rknn.base.RKNNlib.converter.tf_util.TFProto_Util.query_tensor
E File "C:\Users\liufeng\Anaconda3\envs\RKNN_TOOL\lib\site-packages\tensorflow\python\client\session.py", line 887, in run
E    run_metadata_ptr)
E File "C:\Users\liufeng\Anaconda3\envs\RKNN_TOOL\lib\site-packages\tensorflow\python\client\session.py", line 1086, in _run
E    str(subfeed_t.get_shape())))
E ValueError: Cannot feed value of shape (1, 1, 1, 1) for Tensor 'input_holder:0', which has shape '(1,)'
E During handling of the above exception, another exception occurred:
E Traceback (most recent call last):
E File "rknn\api\rknn_base.py", line 331, in rknn.api.rknn_base.RKNNBase.load_tensorflow
E File "rknn\base\RKNNlib\converter\convert_tf.py", line 742, in rknn.base.RKNNlib.converter.convert_tf.convert_tf.match_paragraph_and_param
E File "rknn\api\rknn_log.py", line 312, in rknn.api.rknn_log.RKNNLog.e
E ValueError: Try match Add add failed, catch exception!

只看该作者 · 发表于 2020-7-27 17:54:25

提示不支持tf.add

只看该作者 · 发表于 2020-7-27 18:07:41

import numpy as np

import re
import math
import random
import cv2

from rknn.api import RKNN

INPUT_SIZE = 1

NUM_RESULTS = 1917
NUM_CLASSES = 91

Y_SCALE = 10.0
X_SCALE = 10.0
H_SCALE = 5.0
W_SCALE = 5.0

def expit(x):
return 1. / (1. + math.exp(-x))

def unexpit(y):
return -1.0 * math.log((1.0 / y) - 1.0)

def CalculateOverlap(xmin0, ymin0, xmax0, ymax0, xmin1, ymin1, xmax1, ymax1):
w = max(0.0, min(xmax0, xmax1) - max(xmin0, xmin1))
h = max(0.0, min(ymax0, ymax1) - max(ymin0, ymin1))
i = w * h
u = (xmax0 - xmin0) * (ymax0 - ymin0) + (xmax1 - xmin1) * (ymax1 - ymin1) - i

if u <= 0.0:
      return 0.0

return i / u

def load_box_priors():
box_priors_ = []
fp = open('./box_priors.txt', 'r')
ls = fp.readlines()
for s in ls:
      aList = re.findall('([-+]?\d+(\.\d*)?|\.\d+)([eE][-+]?\d+)?', s)
      for ss in aList:
         aNum = float((ss[0]+ss[2]))
         box_priors_.append(aNum)
fp.close()

box_priors = np.array(box_priors_)
box_priors = box_priors.reshape(4, NUM_RESULTS)

return box_priors

if __name__ == '__main__':

# Create RKNN object
rknn = RKNN()

'''
# Config for Model Input PreProcess
rknn.config(channel_mean_value='127.5 127.5 127.5 127.5', reorder_channel='0 1 2')
'''
# Load TensorFlow Model
print('--> Loading model')
rknn.load_tensorflow(tf_pb='./ts.pb',
                     inputs=['input_holder'],
                     outputs=['predictions'],
                     input_size_list=[[INPUT_SIZE, INPUT_SIZE, 1]])
print('done')

# Build Model
print('--> Building model')
rknn.build(do_quantization=True, dataset='./dataset.txt')
print('done')

# Export RKNN Model
rknn.export_rknn('./ssd_mobilenet_v1_coco.rknn')

只看该作者 · 发表于 2020-7-28 08:51:39

liufeng 发表于 2020-7-27 18:07
import numpy as np

import re

请参照往期帖子中ssd转换例子，或https://github.com/rockchip-linu ... ree/master/examples 中例子，进行转换节点选择。

只看该作者 · 发表于 2020-7-28 10:59:24

leok 发表于 2020-7-28 08:51
请参照往期帖子中ssd转换例子，或https://github.com/rockchip-linux/rknn-toolkit/tree/master/examples ...

好的，谢谢。
我修改了模型，修改输入为通道为3通道的tensor后，再load_tensorflow就成功了，澄清一下可以支持tf.add的操作,其它复杂操作还没测试是否有bug。只是有建议和疑问：
1.rknn只支持固定的3通道tensor，感觉有点限制。
2.如果是支持单通道，rknn需要怎么rknn.config配置。

只看该作者 · 发表于 2020-7-28 11:02:51

if __name__ == '__main__':

# Create RKNN object
rknn = RKNN()

# Config for Model Input PreProcess
rknn.config(channel_mean_value='127.5 127.5 127.5 127.5', reorder_channel='0 1 2')
#rknn.config(channel_mean_value='127.5 127.5', reorder_channel='0')

# Load TensorFlow Model
print('--> Loading model')
rknn.load_tensorflow(tf_pb='./ts.pb',
                     inputs=['input_holder'],
                     outputs=['predictions'],
                     input_size_list=[[INPUT_SIZE, INPUT_SIZE, 3]])
print('done')

这样修改rknn代码，模型代码里输入修改为3通道。成功加载。