colin
/
TIM-VX
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Add quantize, dequantize, requantize test (#232) 

Signed-off-by: Zongwu Yang <zongwu.yang@verisilicon.com>
This commit is contained in:
Zongwu.Yang 2021-12-02 15:40:41 +08:00 committed by GitHub
parent b38bd41933
commit bd496219c8
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 90 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

90
src/tim/vx/ops/simple_operations_test.cc
View File

@ -26,6 +26,7 @@
#include "tim/vx/ops/simple_operations.h"
#include "gtest/gtest.h"
#include <cstdlib>
TEST(Floor, shape_5_1_fp32) {
auto ctx = tim::vx::Context::Create();
@ -83,3 +84,92 @@ TEST(Cast, shape_5_1_fp32_to_int32) {
EXPECT_EQ(golden, output);
}
TEST(DataConvert, quantize_shape_2_3_fp32_to_asym_u8) {
auto ctx = tim::vx::Context::Create();
auto graph = ctx->CreateGraph();
tim::vx::ShapeType io_shape({2, 3});
tim::vx::Quantization quant(tim::vx::QuantType::ASYMMETRIC, 0.0036, 0);
tim::vx::TensorSpec input_spec(tim::vx::DataType::FLOAT32, io_shape,
tim::vx::TensorAttribute::INPUT);
tim::vx::TensorSpec output_spec(tim::vx::DataType::UINT8, io_shape,
tim::vx::TensorAttribute::OUTPUT, quant);
auto input_tensor = graph->CreateTensor(input_spec);
auto output_tensor = graph->CreateTensor(output_spec);
std::vector<float> in_data = {0.8458, 0.6214, 0.4666, 0.6065, 0.8895, 0.1535};
std::vector<uint8_t> golden = {235, 173, 130, 168, 247, 43};
auto quantize = graph->CreateOperation<tim::vx::ops::DataConvert>();
(*quantize).BindInput(input_tensor).BindOutput(output_tensor);
EXPECT_TRUE(graph->Compile());
EXPECT_TRUE(input_tensor->CopyDataToTensor(in_data.data(), in_data.size()*4));
EXPECT_TRUE(graph->Run());
std::vector<uint8_t> output(6, 0);
EXPECT_TRUE(output_tensor->CopyDataFromTensor(output.data()));
EXPECT_EQ(golden, output);
}
TEST(DataConvert, dequantize_shape_2_3_asym_u8_to_fp32) {
auto ctx = tim::vx::Context::Create();
auto graph = ctx->CreateGraph();
tim::vx::ShapeType io_shape({2, 3});
tim::vx::Quantization quant(tim::vx::QuantType::ASYMMETRIC, 0.0036, 0);
tim::vx::TensorSpec output_spec(tim::vx::DataType::FLOAT32, io_shape,
tim::vx::TensorAttribute::OUTPUT);
tim::vx::TensorSpec input_spec(tim::vx::DataType::UINT8, io_shape,
tim::vx::TensorAttribute::INPUT, quant);
auto input_tensor = graph->CreateTensor(input_spec);
auto output_tensor = graph->CreateTensor(output_spec);
std::vector<uint8_t> in_data = {235, 173, 130, 168, 247, 43};
std::vector<float> golden = {0.8458, 0.6214, 0.4666, 0.6065, 0.8895, 0.1535};
auto dequantize = graph->CreateOperation<tim::vx::ops::DataConvert>();
(*dequantize).BindInput(input_tensor).BindOutput(output_tensor);
EXPECT_TRUE(graph->Compile());
EXPECT_TRUE(input_tensor->CopyDataToTensor(in_data.data(), in_data.size()));
EXPECT_TRUE(graph->Run());
std::vector<float> output(6, 0);
EXPECT_TRUE(output_tensor->CopyDataFromTensor(output.data()));
for (uint32_t idx = 0; idx < output.size(); idx++)
EXPECT_TRUE(std::abs(golden[idx] - output[idx]) < 0.01);
}
TEST(DataConvert, requantize_shape_2_3_asym_u8) {
auto ctx = tim::vx::Context::Create();
auto graph = ctx->CreateGraph();
tim::vx::ShapeType io_shape({2, 3});
tim::vx::Quantization in_quant(tim::vx::QuantType::ASYMMETRIC, 0.0036, 0);
tim::vx::Quantization out_quant(tim::vx::QuantType::ASYMMETRIC, 0.0036, 10);
tim::vx::TensorSpec input_spec(tim::vx::DataType::UINT8, io_shape,
tim::vx::TensorAttribute::INPUT, in_quant);
tim::vx::TensorSpec output_spec(tim::vx::DataType::UINT8, io_shape,
tim::vx::TensorAttribute::OUTPUT, out_quant);
auto input_tensor = graph->CreateTensor(input_spec);
auto output_tensor = graph->CreateTensor(output_spec);
std::vector<uint8_t> in_data = {235, 173, 130, 168, 247, 43};
std::vector<uint8_t> golden = {245, 183, 140, 178, 255, 53};
auto requantize = graph->CreateOperation<tim::vx::ops::DataConvert>();
(*requantize).BindInput(input_tensor).BindOutput(output_tensor);
EXPECT_TRUE(graph->Compile());
EXPECT_TRUE(input_tensor->CopyDataToTensor(in_data.data(), in_data.size()));
EXPECT_TRUE(graph->Run());
std::vector<uint8_t> output(6, 0);
EXPECT_TRUE(output_tensor->CopyDataFromTensor(output.data()));
EXPECT_EQ(golden, output);
}