check failed : shift.count == 0 || shift.count == weightCount

Hi,

I try to accelerate a caffe model trained by myself using tensorRT2. And there is a error in a BatchNorm layer:

Parameter check failed at: Network.cpp::addScale::118, condition: shift.count == 0 || shift.count == weightCount
error parsing layer type BatchNorm index 44

Is there some problem with my model parameter?

Thanks!

Hi,

It looks like that there was something wrong with your weight file when TensorRT parsing the batch norm layer.
Which Caffe version do you use? Do you use the FP16 mode?

Thanks.

Hi,

It’s a model based on SSD. And I modify the sampleFasterRCNN when I meet the error. I think it’s the FP16 mode.

Thanks.

Hi,

Not sure which SSD model do you use. Could you share more information with us?

More, could you help us check following commands?

$ ./giexec --deploy=[prototxt] --output=[output blob]
$ ./giexec --deploy=[prototxt] --model=[caffemodel] --output=[output blob]

Thanks.

Hi,

I have solved this problem. There are some problems with a parameter “num_output” in a convolution layer.

Thanks.