[ ERROR ] segment exceeds given buffer limits

[ke__xu]

My PC info:

 Win 10

 Intel i7-7700 3.6 GHz

 8 G Memory

I run a model with name "ResNet_50_cloth_iter_30000", the size of .bin is 91,741 KB = 92 MB

I found that the .bin file for Intel samples are small than 25 MB

When I run my test model, it print as follow:

===log version of 1.2===

[ INFO ] Loading network files for Detection, filename=C:\Intel\computer_vision_sdk_2018.3.343\deployment_tools\model_optimizer\ResNet\ResNet_50_cloth_iter_30000.xml
[ INFO ] Batch size is forced to  1
[ ERROR ] segment exceeds given buffer limits
c:\intel\computer_vision_sdk_2018.3.343\deployment_tools\inference_engine\include\details/ie_exception_conversion.hpp:80
Error 

===Error info of newest version 1.4===

[ ERROR ] metadata is incorrect - segment exceeds given buffer limits. Please validate input data

[nikos1]

Hi Xu,

Interesting! Have never seen this error. I noticed your run on Windows and also using the older version of the SDK

( C:\Intel\computer_vision_sdk_2018.3.343\ )

Could you try 2018 R4?

If it does not work could you try on Linux ?

If you do not have a Linux or an SDK-R4 system available I can try here if you provide link to model and parameters for model optimizer.

Cheers,

Nikos

[nikos1]

> Openvino can run model more than 90MB for GPU? 

Yes, it can.  JFTR my YOLO v3 IR is over 90MB ) and runs fine both on CPU and HD 630 GPU.

FP32: 237M Dec  1 18:32 yolo_v3.bin

FP16 : 119M Dec  1 18:39 yolo_v3.bin

[ke__xu]

Hi Nikos,

 Thank you very much! I will go on try. 

 For version:

 computer_vision_sdk_2018.3.343 error log is:[ ERROR ] segment exceeds given buffer limits

 computer_vision_sdk_2018.4.420 error log is:[ ERROR ] metadata is incorrect - segment exceeds given buffer limits. Please validate input data

Full log for 1.4:

[ INFO ] Loading plugin=GPU
        API version ............ 1.4
        Build .................. 17328
        Description ....... clDNNPlugin
[ MY_LOG ] After Loading plugin=GPU
[ INFO ] Loading network files for Detection, filename=C:\Intel\computer_vision_sdk_2018.4.420\deployment_tools\model_optimizer\ResNet\ResNet_50_cloth_iter_30000.xml
[ INFO ] Batch size is forced to  1
[ ERROR ] metadata is incorrect - segment exceeds given buffer limits. Please validate input data
Error

[nikos1 · Accepted solution]

Thank you Xu!

Error may be related to your network. How did you freeze? Can you repro this with any pretrained networks available in public?

JFTR the error code seems to be in https://github.com/opencv/dldt/blob/55a41d7570f78aaea0d6764d157dd7434730d56f/inference-engine/src/inference_engine/v2_format_parser.cpp

template<typename BlobType>
inline Blob::Ptr GetTypedBlobFromSegment(const TBlob<uint8_t>::Ptr& weights, const WeightSegment& segment) {
    if (segment.getEnd() > weights->size())
        THROW_IE_EXCEPTION << "metadata is incorrect - segment exceeds given buffer limits. Please validate input data";

Does this give you any clues?

Cheers,

Nikos

[ke__xu]

Hi Nikos,

 Thank you very much! It is something wrong when copy input bin file.

Best Regards,

Ke