Intel® Distribution of OpenVINO™ Toolkit
Community support and discussions about the Intel® Distribution of OpenVINO™ toolkit, OpenCV, and all things computer vision-related on Intel® platforms.

NCSDK 2.10 OPTION: --tf-ssd-config

Community Manager

The --tf-ssd-config option allows the TensorFlow SSD MobileNet v1/v2 coco networks to be used with the NCSDK 2.10 toolkit commands: mvNCCompile, mvNCCheck, and mvNCProfile.



NCSDK TensorFlow ssd config file format:


TensorFlow SSD config files for NCSDK follow a simple template. Modifications to your config file may be necessary for compatibility with the NCSDK.


Example TensorFlow ssd config file:


input_height: 300 input_width: 300 postprocessing_params { num_classes: 91 background_label_id: 0 max_detections: 100 nms_params { score_threshold: 0.300000011921 iou_threshold: 0.600000023842 max_detections_per_class: 100 } box_params { var: 0.1 var: 0.1 var: 0.2 var: 0.2 } } score_converter: SIGMOID





This option can be used with the mvNCCompile, mvNCCheck and mvNCProfile toolkit commands. The usage is as follows:





Example mvNCCompile command:


mvNCCompile tf_ssd_mobilenet_v1.pb --tf-ssd-config ncsdk_tf_ssd.config -s 12
0 Kudos
5 Replies
Community Manager

Hi @Jesus_at_Intel,


We trained for traffic signs of 47classes using ssd_inception_v2 and tensorflow object detection API.


When we try to convert into NCS graph, No error and after 1-2min machine is going into NO RESPOND mode(stuck) .


Can you help us to resolve the issue?




Sai Krishna.
Community Manager

After trying this, I get the following error:


[Error 13] Toolkit Error: Provided OutputNode/InputNode name does not exist or does not match with one contained in model file Provided: . Input node flag not supported for SSD networks.


when removing the input node flag I get:


[Error 13] Toolkit Error: Provided OutputNode/InputNode name does not exist or does not match with one contained in model file Provided: . Could not infer input node name.
Community Manager

Hi @saikrishnaTheGreat ,


How did you set your config file? I have a SSDLite_MobileNet_v2_coco model which has been trained on my own dataset and has only one class to detect. It works fine in the form of TF frozen model, but after transforming it to NCS graph, it cannot output anything.


In my case, I just change the num_classes from 90 to 1 in the ncsdk_tf_ssd.config given by @Jesus_at_Intel . However, I have no confidence in my config file settings.


By the way, I can convert the default frozen model of SSDLite_MobileNet_v2_coco (which has 90 object classes) by following @Jesus_at_Intel 's example and it works as expected.

Community Manager

I also try to set num_classes to 2, and the output is some almost-static bounding boxes with nonsense high class scores. (The following screenshot is the output in which the label name is irrelevant.)


Community Manager

I used the ncsdk_tf_ssd.config but the system (mobilentSSD_v1) does not detect much at all. Ran same model on the CPU and detected objects fine. The config file from the model has the following information {


ssd {


num_classes: 90


image_resizer {


fixed_shape_resizer {


height: 300


width: 300




So I change the num_classes in ncsdk_tf_ssd.config to 90 and similar result to the above. Any thoughts?