Exporting TfLite QBatch Permutation

[sciapponi]

In micromind.core.export()

qbatch is an optional function argument and it's assigned a None value if not provided.

Before calling micromind.convert_to_tflite() at line 297, qbatch is permuted as a four dimensional tensor from (B, C, W, H) to (B,H,W,C) which doesn't generalize to different input sizes or deal with None Input.

If qbatch is omitted:

m.export("tflite", "tflite", [1,100,777])

the raised error is:

Traceback (most recent call last):
  File "/home/ste/Code/python/tiny-vocos/test_xinet.py", line 597, in <module>
    m.export("tflite", "tflite", [1,100,777])
  File "/home/ste/Code/python/micromind/micromind/core.py", line 297, in export
    qbatch = qbatch.permute(0, 3, 2, 1)
AttributeError: 'NoneType' object has no attribute 'permute'

To fix this we would need a general permutation implementation for any dimension and a statement which deals with NoneType input.