How can I get the individual yolov3_output layer's NDArray from the TVM Framewrok?

Questions
1

I tried to run the YOLOV3 network via the TVM Framework, and using the related YOLOV3’s cfg/weights.

For running the “/tvm/tutorials/nnvm/from_darknet.py” script, it works, but the result is not correct:

The yolov3 has three outputs - YOLO layer82/layer94/layer106 for the next postprocess, but the TVM framework detected only one output - layer106.
So, how can I get the layer82/94’s ndarry via the TVM Framwork for the next post process ?
Like using the API: tvm_out = m.get_output(0).asnumpy().flatten()

The script “from_darknet.py” can compile&run YOLOV2 successfully. But seemingly, there is a difference between yolov2’s postprocess and yolov3’s postprocess.
If one day, the TVM framework can privide yolov3’s postprocess that would be thankful.

2

Current darknet frontend script only output last layer output of yolo models. Yes for yolov3, we do need intermediate layer outputs. I will work on your suggested modifications soon. Thanks.