Calibration Tool Object detection Error

[V_S__Suchithra]

Hi,

I would like to convert FP32 object detection model to int8 model and I tried to convert the pascal model as per the documentation. While doing this, some error occured.

Below I am giving the command which i used to convert from fp32 model to int8 model.

./calibration_tool -t OD -d CPU -i ./Desktop/Pascal_Dataset/VOCdevkit_Test -ODa ./Desktop/Pascal_Dataset/VOCdevkit_Test -m ./Desktop/Pascal_Dataset/VGG16_faster_rcnn_final.xml -ODc ./Desktop/Pascal_Dataset/pascal_voc_classes.txt -ODsubdir JPEGImages -subset 1000

Below the details of the error.

[ INFO ] InferenceEngine:
    API version ............ 1.6
    Build .................. custom_releases/2019/R1.1_28dfbfdd28954c4dfd2f94403dd8dfc1f411038b
[ INFO ] Parsing input parameters
[ INFO ] Loading plugin

    API version ............ 1.6
    Build .................. 23780
    Description ....... MKLDNNPlugin
[ INFO ] Loading network files
[ INFO ] Preparing input blobs
[ ERROR ] Inference problem:
This app accepts networks having only one input

I am using Openvino 2019 R1 version.

Could you please tell me why I am getting this error?

[Shubha_R_Intel]

Dear V S, Suchithra,

First you are using too old of an OpenVino version. Please upgrade to OpenVino 2019R2 which was recently released within the last 2 weeks. Also, I have given detailed step-by-step instructions on how to use the Python Calibration tools in dldt github response 171 . There are many iterations involved.

Please post back here should you run into difficulties.

Thanks !

Shubha

[V_S__Suchithra]

Hi Shubha,

Thank you for your information.

[Shubha_R_Intel]

Dear V S, Suchithra,

Of course. And we have also released a Preview Version of DL Workbench in OpenVino R2. In the future it will make calibration of INT8 and WinoGrad optimized models much smoother and easier.

Shubha

[V_S__Suchithra]

Hi Shubha,

I am using OpenVino R2 version.

When I tried to execute convert_annotation.py as the first step, there occurs some error which tells that there is an argument converter is needed. But for my custom dataset, the annotation converter is not supported. How can i solve this problem?

Error :

(choose from 'voc_detection', 'movie_lens_converter', 'imagenet', 'detection_opencv_storage', 'brats', 'sample', 'mscoco_keypoints', 'icdar13_recognition', 'mscoco_segmentation', 'cifar10', 'voc_segmentation', 'wider', 'brats_numpy', 'vgg_face', 'super_resolution', 'mscoco_mask_rcnn', 'mnist_csv', 'lfw', 'market1501_reid', 'mscoco_detection', 'icdar15_detection', 'mars', 'cityscapes')

Thanks & Regards

Suchithra

[Shubha_R_Intel]

Dear V S, Suchithra

If your custom dataset is not reflected in the below directory then you need to make yourself a custom annotation_converter. That is the only way. PS, your python version may be different.

 C:\Program Files (x86)\IntelSWTools\openvino_2019.2.242\python\python3.7\openvino\tools\accuracy_checker\accuracy_checker\annotation_converters

Thanks,

Shubha