Hi,
I use the SSD_Mobilenet DLC to produce the 3 outputs with snpe-dlc-run, which caused by add the "#Postprocessor/BatchMultiClassNonMaxSuppression add_6" in the first line of the "raw_list.txt".
However, when I use the same SSD_Mobilenet DLC with the NativeCpp builded snpe-sample.
The output is only detection_classes.
I manually changed the function createOutputBufferMap in CreateUserBuffer.cpp.
const char *name = "detection_boxes:0"; or "detection_boxes" or "Postprocessor/BatchMultiClassNonMaxSuppression_boxes" or "Postprocessor/BatchMultiClassNonMaxSuppression"
createUserBuffer(outputMap, applicationBuffers, snpeUserBackedBuffers, snpe, name);
the terminate called after throwing an instance of 'std::runtime_error'what(): Error obtaining attributes for input tensor detection_boxesAborted (core dumped)
Hi hsiaoyuh_wang,
I am facing the same issue. Were you able to solve it.
Hi all,
I got the same issue with you guys. The converter not worked very well for mobilenet-ssd v1. if you add --verbose as an option, you will find only detection_class setted as output.
I tried to understanding the detail of converter, the expected output node such as detection_score removed by function "_resolve_hierarchical_resolution_conflicts", I still don't know why. Could you guys share your ideas to fix this?
BTW, do you guys have a try to resize the network's input from 4,300,300,3 to another shape? I cannot make this resize.
Thanks,
Hi,
I've regerated the DLC and then ran the snpe-dlc-run with raw_list.txt file having
#Postprocessor/BatchMultiClassNonMaxSuppression add_6
as the first line. Then it worked. Got
Postprocessor/BatchMultiClassNonMaxSuppression_classes.raw
Postprocessor/BatchMultiClassNonMaxSuppression_scores.raw
Postprocessor/BatchMultiClassNonMaxSuppression_boxes.raw
in the output folder for each of the image input.
But still don't know the actual reason for not working in the first try.