Toybrick

标题: vgg16-bn.onnx不开量化预测成功，开量化build失败！ [打印本页]

作者: shanshuiziran 时间: 2019-6-27 23:52
标题: vgg16-bn.onnx不开量化预测成功，开量化build失败！
vgg16-bn.onnx不开量化转换成功，预测也成功；开量化build失败！同样的转换代码，转另一个onnx 就可以成功，不知什么原因？
转换代码：
#_*_coding: utf-8 _*_
import numpy as np
import cv2
from rknn.api import RKNN
from PIL import Image

if __name__ == '__main__':
# Create RKNN object
rknn = RKNN(verbose=True,verbose_file='./crnntcn.log')

# pre-process config
print('--> config model')
rknn.config( channel_mean_value='128 128 128 128', reorder_channel='0 1 2')#,batch_size=1,epochs=500#batch_size=1, ,quantized_dtype='dynamic_fixed_point-16',quantized_dtype='dynamic_fixed_point-8'
print('done')

# Load tensorflow model
print('--> Loading model')
ret = rknn.load_onnx(model = './vgg16-bn.onnx')
if ret != 0:
      print('Load crnn_cnn failed! Ret = {}'.format(ret))
      exit(ret)
print('done')

# Build model
print('--> Building model')
ret = rknn.build(do_quantization=True, dataset='/home/sysman/rknn/mv2/vggpic/dataset.txt')
if ret != 0:
      print('Build crnn_cnn failed!')
      exit(ret)
print('done')

# Export rknn model
print('--> Export RKNN model')
ret = rknn.export_rknn('./vgg.rknn')
if ret != 0:
      print('Export crnn_cnn.rknn failed!')
      exit(ret)
print('doneOKOK')

rknn.release()

出错部分：
。。。。。
D Quantize tensor @Conv_17_37:bias.
D Quantize tensor @Conv_14_40:bias.
D Quantize tensor @Conv_10_44:bias.
D Quantize tensor @Conv_7_47:bias.
D Quantize tensor @Conv_3_51:bias.
D Quantize tensor @Conv_0_54:bias.
I Clean.
D Optimizing network with qnt_single_transmit_quantize, align_quantize, broadcast_quantize, qnt_adjust_coef
D Quantize tensor(@MaxPool_43_11

ut0) with tensor(@Relu_42_12

ut0)
D Quantize tensor(@attach_Gemm_53/out0_0

ut0) with tensor(@Gemm_53_1

ut0)
[TRAINER]Quantization complete.
E Catch exception when building RKNN model!
T Traceback (most recent call last):
T File "rknn/api/rknn_base.py", line 498, in rknn.api.rknn_base.RKNNBase.build
T File "rknn/api/rknn_base.py", line 1071, in rknn.api.rknn_base.RKNNBase._quantize
T File "rknn/base/RKNNlib/app/tensorzone/quantization.py", line 151, in rknn.base.RKNNlib.app.tensorzone.quantization.Quantization.run
T File "rknn/base/RKNNlib/RKNNnet.py", line 341, in rknn.base.RKNNlib.RKNNnet.RKNNNet.save_data
T File "rknn/base/RKNNlib/RKNNnet.py", line 355, in rknn.base.RKNNlib.RKNNnet.RKNNNet._save_old_data
T File "rknn/base/RKNNlib/RKNNnet.py", line 363, in rknn.base.RKNNlib.RKNNnet.RKNNNet._save_old_data
T File "/usr/local/lib/python3.5/dist-packages/numpy/lib/npyio.py", line 511, in save
T pickle_kwargs=pickle_kwargs)
T File "/usr/local/lib/python3.5/dist-packages/numpy/lib/format.py", line 586, in write_array
T pickle.dump(array, fp, protocol=2, **pickle_kwargs)
T MemoryError

请高手指点，多谢了

作者: shanshuiziran 时间: 2019-7-1 18:08
希望官方回复！！！

作者: jefferyzhang 时间: 2019-7-4 09:19
查看最新的rknn toolkit的文档，新版rknn（1.0版本以及未来1.0以上版本）都支持混合量化，可以逐层量化预测结果，查看哪一层量化会出问题，然后单独关闭那一层的量化。实现高度自由配置。

作者: zhangzj 时间: 2019-7-15 09:11

shanshuiziran 发表于 2019-7-1 18:08
希望官方回复！！！

T MemoryError

是不是这个网络模型比较大，爆内存了？量化会比较消耗内存。

欢迎光临 Toybrick (https://t.rock-chips.com/)