TDA4VMXEVM: [TIDL] import caffe model to tidl model error

Nigel Peng

Part Number: TDA4VMXEVM

Dear TI,

Now I have downlaod a public model (alexnet) from github, you know it's a simple & small model.but import error,

there is a un-support layer LRN,but some other error I just don't understand, the log is as below:

Caffe Network File : ../../test/testvecs/models/public/caffe/alexnet/bvlc_alexnet_deploy.prototxt
Caffe Model File : ../../test/testvecs/models/public/caffe/alexnet/bvlc_alexnet_no_group.caffemodel
TIDL Network File : ../../test/testvecs/config/tidl_models/caffe/tidl_net_bvlc_alexnet_no_group.caffemodel.bin
TIDL IO Info File : ../../test/testvecs/config/tidl_models/caffe/tidl_io_bvlc_alexnet_no_group.caffemodel_
Name of the Network : AlexNet
WARNING: detect batch process from input size, overwrite to single frame inference!
ERROR: LRN layer norm2 is not suported now.. By passing
WARNING: !!!!!!! Not supported layer LRN:norm2 is found! Import result is unpredictable!
ERROR: LRN layer norm1 is not suported now.. By passing
WARNING: !!!!!!! Not supported layer LRN:norm1 is found! Import result is unpredictable!
WARNING: Conv Layer conv2's coeff cannot be found(or not match) in coef file, Random coeff will be generated! Only for evaluation usage! Results are all random!
WARNING: Conv Layer conv4's coeff cannot be found(or not match) in coef file, Random coeff will be generated! Only for evaluation usage! Results are all random!
WARNING: Conv Layer conv5's coeff cannot be found(or not match) in coef file, Random coeff will be generated! Only for evaluation usage! Results are all random!
In put of TIDL_InnerProductLayer layer needs to be Faltten. Please add Flatten layer to import this mdoel

I have 2 questions

1、the conv2's coeff can't be found,I have add some log before the err line

printf("[%s] %#x,bufSize=%d,size=%d\n",TIDLPCLayers.name, TIDLPCLayers.weights.ptr, TIDLPCLayers.weights.bufSize, dataSize);
if(TIDLPCLayers.weights.ptr == NULL ||
TIDLPCLayers.weights.bufSize != dataSize)
{
printf("WARNING: Conv Layer %s's coeff cannot be found(or not match) in coef file, "
"Random coeff will be generated! "
"Only for evaluation usage! "
"Results are all random!\n", TIDLPCLayers.name);

...

}

the log after import is like this [conv2] 0xd281010,bufSize=614400,size=307200,

so the condition doesn't match,please tell me where is the param from ,is from the prototxt? is there any document can refer to.

2、as the err log "TIDL_InnerProductLayer layer needs to be Faltten." should I only add this layer before TIDL_InnerProductLayer in prototxt ?and the caffe model don't have any change?

TKS

Nigel

over 5 years ago

0 kumar.desappan over 5 years ago

TI__Mastermind 22145 points

Hi, Nigel,

LRN layer is not supported in TIDL, So Alex net import is failing.

Please refer the below for 2o+ Validated public models in TIDL.

http://software-dl.ti.com/jacinto7/esd/processor-sdk-rtos-jacinto7/latest/exports/docs/tidl_j7_01_01_00_10/ti_dl/docs/user_guide_html/md_tidl_models_info.html

0 Nigel Peng over 5 years ago in reply to kumar.desappan

Intellectual 525 points

Kumar

I have konw LRN is un-support layer yet,so I want to add this layer to custom layer.

so I asked 2 questions.

Nigel

0 Ming Zhong10 over 5 years ago in reply to Nigel Peng

TI__Intellectual 2235 points

Hi,

1. Your prototxt is not match with your caffemodel, so import tool cannot find correct param from your caffemodel.

2. you only need to modify your prototxt. add a flatten layer before inner product layer.

Thanks & Best Regards!

0 Nigel Peng over 5 years ago in reply to Ming Zhong10

Intellectual 525 points

Hi Ming,

I have replace this model with a no group caffemodel.now the log is

so as you suggest add a flatten before inner product layer,the prototxt and import.cfg as attachment.

it's seem generate the bin success,but there are other errors as below:

ERROR: LRN layer norm2 is not suported now.. By passing
WARNING: !!!!!!! Not supported layer LRN:norm2 is found! Import result is unpredictable!
ERROR: LRN layer norm1 is not suported now.. By passing
WARNING: !!!!!!! Not supported layer LRN:norm1 is found! Import result is unpredictable!
Could not find Indata for data ID 2
Could not find Indata for data ID 5

~~~~~Running TIDL in PC emulation mode to collect Activations range for each layer~~~~~

Processing config file #0 : /home/work/psdk_rtos_auto_j7_06_01_01_12/tidl_j7_01_00_01_00/ti_dl/utils/tidlModelImport/tempDir/qunat_stats_config.txt
----------------------- TIDL Process with REF_ONLY FLOW------------------------

OpenCV Error: Assertion failed (0 <= roi.x && 0 <= roi.width && roi.x + roi.width <= m.cols && 0 <= roi.y && 0 <= roi.height && roi.y + roi.height <= m.rows) in Mat, file /home/work/ti_tools/opencv-3.1.0/modules/core/src/matrix.cpp, line 508
terminate called after throwing an instance of 'cv::Exception'
what(): /home/work/ti_tools/opencv-3.1.0/modules/core/src/matrix.cpp:508: error: (-215) 0 <= roi.x && 0 <= roi.width && roi.x + roi.width <= m.cols && 0 <= roi.y && 0 <= roi.height && roi.y + roi.height <= m.rows in function Mat

Aborted (core dumped)

is the err log happen cause by LRN does't add to custom layer? in my opinion，the LRN is bypass,so I think it's should be runing ok.

do you have any doc can guide there steps.TKS.

BR.

0 Nigel Peng over 5 years ago in reply to Nigel Peng

Intellectual 525 points

the tidl import is as below,it's modify from mobilenetv2.txt

modelType = 0
numParamBits = 8
inputNetFile = "../../test/testvecs/models/public/caffe/alexnet/bvlc_alexnet_no_group_deploy.prototxt"
inputParamsFile = "../../test/testvecs/models/public/caffe/alexnet/bvlc_alexnet_no_group.caffemodel"
outputNetFile = "../../test/testvecs/config/tidl_models/caffe/alexnet/tidl_net_alexnet.bin"
outputParamsFile = "../../test/testvecs/config/tidl_models/caffe/alexnet/tidl_io_alexnet_"
inDataNorm = 1
inMean = 103.94, 116.78, 123.68
inScale = 0.017 0.017 0.017
inDataFormat = 0
resizeWidth = 224
resizeHeight = 224
inNumChannels = 3
inData = ../../test/testvecs/config/imageNet_sample_val.txt
postProcType = 1

the prototxt add a flatten layer and modify as below

layer {
name: "conf_flat"
type: "Flatten"
bottom: "pool5"
top: "conf_flat"
flatten_param {
axis: 1
}
}

layer {
name: "fc6"
type: "InnerProduct"
bottom: "conf_flat"
top: "fc6"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
inner_product_param {
num_output: 4096
}
}

0 Ming Zhong10 over 5 years ago in reply to Nigel Peng

TI__Intellectual 2235 points

Per discussed through mail. Can we close this issue?

You should remove all unsupported layer before import. OR add unsupported layer as custom layer in import tool before import.

Processors

Processors forum

TDA4VMXEVM: [TIDL] import caffe model to tidl model error