Hi,
I built the tensorflow lite on iMX8(4xA53) platform. I can get correct results when using models of mobilenet_v1_1.0_224, mobilenet_v1_1.0_224_quant and mobilenet_v2_1.0_224 as input model for label_image. But I got unreasonable predictons when using mobilenet_v2_1.0_quant.

Here are the results:
./label_image -i dog.bmp -m mobilenet_v1_1.0_224.tflite -l labels.txt
Loaded model mobilenet_v1_1.0_224.tflite
resolved reporter
invoked
average time: 200.31 ms
0.987781: 209 Labrador retriever
0.00432148: 208 golden retriever
0.00298653: 163 beagle
0.00134322: 160 Rhodesian ridgeback
./label_image -i dog.bmp -m mobilenet_v1_1.0_224_quant.tflite -l labels.txt
Loaded model mobilenet_v1_1.0_224_quant.tflite
resolved reporter
invoked
average time: 85.934 ms
0.984314: 209 Labrador retriever
0.00392157: 435 bath towel
0.00392157: 208 golden retriever
0.00392157: 169 redbone
0.00392157: 163 beagle
./label_image -i dog.bmp -m mobilenet_v2_1.0_224.tflite -l labels.txt
Loaded model mobilenet_v2_1.0_224.tflite
resolved reporter
invoked
average time: 180.721 ms
0.946971: 209 Labrador retriever
0.00624463: 160 Rhodesian ridgeback
0.0035156: 208 golden retriever
0.00168908: 244 bull mastiff
0.0015452: 163 beagle
./label_image -i dog.bmp -m mobilenet_v2_1.0_224_quant.tflite -l labels.txt
Loaded model mobilenet_v2_1.0_224_quant.tflite
resolved reporter
invoked
average time: 87.001 ms
0.780392: 209 Labrador retriever
0.529412: 853 tennis ball
0.529412: 160 Rhodesian ridgeback
0.509804: 208 golden retriever
0.498039: 244 bull mastiff

My questionn is that is this a bug about the Mobilenet_V2 quantized models in tensorflow lite 2.0 alpha? Would you please help? Thanks in advance.