diff options
| author | ricbur01 <richard.burton@arm.com> | 2019-10-31 16:22:18 +0000 |
|---|---|---|
| committer | Matteo Martincigh <matteo.martincigh@arm.com> | 2019-11-01 08:38:16 +0000 |
| commit | 3f4d7104cdd5fd0ca6238afd4d22ac8734fd84ae (patch) | |
| tree | 69cbc14e7216f0d5a28560a60fab5e735e03c51c | |
| parent | 9a5f9f2c3b6709003dfb935ecf65150145b9c73b (diff) | |
| download | armnn-3f4d7104cdd5fd0ca6238afd4d22ac8734fd84ae.tar.gz | |
Increase tolerance for bias scale quantization calculation for ResNet50v2
Signed-off-by: Richard Burton <richard.burton@arm.com>
Change-Id: I033ef445fa8e5ae825cfde29927b88e97e20f093
| -rw-r--r-- | src/backends/backendsCommon/WorkloadData.cpp | 2 |
1 files changed, 1 insertions, 1 deletions
diff --git a/src/backends/backendsCommon/WorkloadData.cpp b/src/backends/backendsCommon/WorkloadData.cpp index e3d78fea50..e1a369af7c 100644 --- a/src/backends/backendsCommon/WorkloadData.cpp +++ b/src/backends/backendsCommon/WorkloadData.cpp @@ -195,7 +195,7 @@ void ValidateBiasTensorQuantization(const TensorInfo& biasTensor, // Helper lambda function to validate a single bias quantization scale value auto VerifyBiasQuantizationScale = [&descName](float biasScale, float expectedScale) -> void { - constexpr float tolerance = 0.00000001f; + constexpr float tolerance = 0.000001f; if (std::abs(biasScale - expectedScale) > tolerance) { // Print the float values with extra precision to see very small differences |