Add fix for MobileNetSSD Bug on Android

[swarmidentity]

Seems like TFLite conversion on TensorFlow 2.7.0 acts differently than previous versions - when I run the python script:

`
num_classes=1
tflite_interpreter.resize_tensor_input(input_details[0]['index'], (1, 320, 320, 3))
tflite_interpreter.resize_tensor_input(output_details[0]['index'], (1, num_classes))
tflite_interpreter.allocate_tensors()

input_details = tflite_interpreter.get_input_details()
output_details = tflite_interpreter.get_output_details()

print("== Input details ==")
print("shape:", input_details[0]['shape'])
print("\n== Output details ==")
print("shape:", output_details[0]['shape'])
print("shape:", output_details[1]['shape'])
print("shape:", output_details[2]['shape'])
print("shape:", output_details[3]['shape'])

print("name:", output_details[0]['name'])
print("name:", output_details[1]['name'])
print("name:", output_details[2]['name'])
print("name:", output_details[3]['name'])
`

on a TFLite file generated a few months ago, I get:
`
== Input details ==
shape: [ 1 320 320 3]

== Output details ==
shape: [ 1 10 4]
shape: [ 1 10]
shape: [ 1 10]
shape: [1]

name: StatefulPartitionedCall:3
name: StatefulPartitionedCall:2
name: StatefulPartitionedCall:1
name: StatefulPartitionedCall:0
`

On a TFLite file generated with 2.7.0, I get:
`
== Input details ==
shape: [ 1 320 320 3]

== Output details ==
shape: [ 1 10]
shape: [ 1 10 4]
shape: [1]
shape: [ 1 10]

name: StatefulPartitionedCall:1
name: StatefulPartitionedCall:3
name: StatefulPartitionedCall:0
name: StatefulPartitionedCall:2
`

This patch fixes the specific error I'm running into on Android for MobileNetSSD models, there may be other cases that have similar issues like #223