[solved]Getting same output after deploying tflite on arm_cpu

Questions
1

Hi, I used (https://github.com/dmlc/nnvm/blob/master/docs/how_to/deploy.md) for building an application to run on android device. I am allocating random values to the tvm array x and calling the set_input function(), but everytime I get the same output, I mean the same index for max value out of 1000 values and same values per index. The case is same with the tuned as well as untuned graph. And I get a continuous warning saying: /home/yovan/tvm_all/run_tvm_on_device/tvm/src/runtime/graph/graph_runtime.cc:65: Warning: cannot find “data” among input . Output is something like this (for tflite mobilenet_v1_224):
0.000127
0.000527
0.000987
0.000252
0.000207
0.000938
0.000505
0.000070
0.000158
0.000119
0.000649
0.000668
0.000274
0.000112
0.000165
0.000156
0.000298
0.000191
0.000221
0.000192
0.000126
0.000267
0.001076
0.001035
0.000715
0.000360…
…
Note: there should be 2 changes in the above code

  1. int out_ndim = 2;
  2. int64_t out_shape[] = {1, 1001};
2

The input tensor is not data, is input.

3

Worked, thanks! but is is not just a naming convention? Like, we are providing a name ‘data’ or ‘input’ to the tensor, or the name we had provided in the python file for the graph it should match with that?

4

This name is the input name of model. I think I am anwering the same question secondly for you. I must say that you should follow the link mentioned by my first answer of your same question: Problem in tuning and deploying tflite model on arm target via rpc

5

yeah, sorry for that, but how come some inference was coming from after application was called, it does it have some default inputs?

6

Do you mean the image data for input? for example cat.png or else?

7

Yes, the image data for input, when I was assigning using set_input(“data”, x) x being the TVMArrayAlloc, it could’t find the name for the input tensor defined in the library right?
So, how come inference could come?