What are the changes need to make in code to pass Tensorflow MobileNetSSD model(mobilenet_ssd.dlc)

[sainisanjay]

Hi @rakesh-sankar, App works good, but when i use mobilenet_ssd.dlc it does not detect any object. I guess mobilenet_ssd.dlc input node in Preprocessor/sub and output node is in Postprocessor/BatchMultiClassNonMaxSuppression_boxes Postprocessor/BatchMultiClassNonMaxSuppression_scores Postprocessor/BatchMultiClassNonMaxSuppression_classe But in app code passes only one input and output node which is true in caffe_mobilenet.dlc and object_detect.dlc.

private static final String MNETSSD_INPUT_LAYER = "data";
private static final String MNETSSD_OUTPUT_LAYER = "detection_out";

Can any one tell me how to pass input and output node for tensorflow based mobilenet_ssd.dlc which as multiple output nodes.

[rakesh-sankar]

Yeah, True! We have used the MobileNetSSD Model which has built on top of caffe framework, Model with Caffe framework is having only 1 Input node and 1 output node.

For Tensorflow based MobileNet SSD Model, Need to be converted by mentioning the 3 output node names. Find command given below for conversion: snpe-tensorflow-to-dlc --input_network <path_to>/exported/frozen_inference_graph.pb --input_dim Preprocessor/sub 1,300,300,3 --out_node detection_classes --out_node detection_boxes --out_node detection_scores ---output_path mobilenet_ssd.dlc --allow_unconsumed_nodes

After SNPE conversion you should have a mobilenet_ssd.dlc that can be loaded and run in the SNPE runtimes.

The output layers for the model are:

    Postprocessor/BatchMultiClassNonMaxSuppression
    add

The output buffer names are:

    (classes) detection_classes:0 (+1 index offset)
    (classes) Postprocessor/BatchMultiClassNonMaxSuppression_classes (0 index offset)
    (boxes) Postprocessor/BatchMultiClassNonMaxSuppression_boxes
    (scores) Postprocessor/BatchMultiClassNonMaxSuppression_scores

You have to do required changes in code to get the output from the multiple nodes so you can map them out easily.