IVGCVSW-3129 Image pre-processing fix for TFLite
authorFinnWilliamsArm <Finn.Williams@arm.com>
Wed, 22 May 2019 13:50:55 +0000 (14:50 +0100)
committerNarumol Prangnawarat <narumol.prangnawarat@arm.com>
Mon, 27 May 2019 16:27:36 +0000 (16:27 +0000)
commitaf8b72d3a97baa0c86583b03321e8a8d3f4bc2aa
tree3b4d86842399842b852e593c3bcf131d86d3ae20
parent43aec5886449c1b024b740fd6f4500e827bde221
IVGCVSW-3129 Image pre-processing fix for TFLite

 * Resized images for quantized models are now statically cast to uint8
   instead of quantized

 * Removed optional quantization parameters from ImagePreprocessor
   constructor

 * Changed mean and scale for TFLite models

Signed-off-by: FinnWilliamsArm <Finn.Williams@arm.com>
Change-Id: Id5ffdf77f3614d10c417e769bd8ffc4a4c07308b
14 files changed:
tests/ImagePreprocessor.cpp
tests/ImagePreprocessor.hpp
tests/InferenceTestImage.cpp
tests/InferenceTestImage.hpp
tests/OnnxMobileNet-Armnn/OnnxMobileNet-Armnn.cpp
tests/TfLiteInceptionV3Quantized-Armnn/TfLiteInceptionV3Quantized-Armnn.cpp
tests/TfLiteInceptionV4Quantized-Armnn/TfLiteInceptionV4Quantized-Armnn.cpp
tests/TfLiteMnasNet-Armnn/TfLiteMnasNet-Armnn.cpp
tests/TfLiteMobileNetQuantizedSoftmax-Armnn/TfLiteMobileNetQuantizedSoftmax-Armnn.cpp
tests/TfLiteMobilenetQuantized-Armnn/TfLiteMobilenetQuantized-Armnn.cpp
tests/TfLiteMobilenetV2Quantized-Armnn/TfLiteMobilenetV2Quantized-Armnn.cpp
tests/TfLiteResNetV2-50-Quantized-Armnn/TfLiteResNetV2-50-Quantized-Armnn.cpp
tests/TfLiteResNetV2-Armnn/TfLiteResNetV2-Armnn.cpp
tests/TfLiteVGG16Quantized-Armnn/TfLiteVGG16Quantized-Armnn.cpp