tensorflow: first_bn/FusedBatchNorm_mul_0, which is an input to the Add operator producing the output array kws_model/KWS_Model/tower_0/CNN_V1/Relu, is lacking min/max data, which is necessary for quantization. Either target a non-quantized output format, or change the input graph to contain min/max information, or pass --default_ranges_min= and --default_ranges_max= if you do not care about the accuracy of results.\nAborted (core dumped)"
System information -Centos
- TensorFlow 1.9:
**change the pb model to tflite * my network include batch_normalization layer, when I change the the pb model to tflite , there is always return a error message.
first_bn/FusedBatchNorm_mul_0, which is an input to the Add operator producing the output array kws_model/KWS_Model/tower_0/CNN_V1/Relu, is lacking min/max data, which is necessary for quantization. Either target a non-quantized output format, or change the input graph to contain min/max information, or pass --default_ranges_min= and --default_ranges_max= if you do not care about the accuracy of results.\nAborted (core dumped)"
I tried lots of methods, like add converter.default_ranges_stats=(0, 6) or remove bn layers. but I know its not the expected ways. hope other friends can give me some suggestions. thanks
About this issue
- Original URL
- State: closed
- Created 5 years ago
- Comments: 18 (1 by maintainers)
@achandraa Setting default ranges for ops with missing min/max can’t possibly be an acceptable solution.
The error message clearly states that it should only be used if accuracy isn’t important (in what scenario would this be true other than for experimentation?):
Either target a non-quantized output format, or change the input graph to contain min/max information, or pass --default_ranges_min= and --default_ranges_max= if you do not care about the accuracy of results.