update readme

Author: yester31

README.md

@@ -85,15 +85,16 @@
 
 ## Super-Resolution model(in progress)
 - TensorRT 8.0.3.4 (Real-ESRGAN) 
-- Real-ESRGAN model (real-esrgan.cpp) 
+- Real-ESRGAN model (real-esrgan.cpp)
+- Scale up 4x (448x640x3 -> 1792x2560x3) 
 - Comparison of calculation execution time of 100 iteration and GPU memory usage for one 448x640x3
-  - Pytorch  F32	
-  - Pytorch  F16	
-  - TensorRT F32	
-  - TensorRT F16	
+  - Pytorch  F32	4109 ms ( 5.029 GB)
+  - Pytorch  F16	1936 ms ( 4.407 GB)
+  - TensorRT F32	2139 ms ( 3.807 GB) (0.47 FPS)
+  - TensorRT F16	737 ms ( 3.311 GB) (1.35 FPS)
   - TensorRT Int8	
 ***
-
+ 
 ## Using C TensoRT model in Python using dll
 - TRT_DLL_EX : <https://github.com/yester31/TRT_DLL_EX>
 ***

Real-ESRGAN_py/results/OST_009_out.png

@@ -135,6 +135,7 @@
     </ClCompile>
     <ClCompile Include="yolov5s.cpp">
       <ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">true</ExcludedFromBuild>
+      <ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Release|x64'">true</ExcludedFromBuild>
     </ClCompile>
   </ItemGroup>
   <ItemGroup>

TensorRT/TensorRT.vcxproj

@@ -143,7 +143,6 @@ ITensor* residualDenseBlock(INetworkDefinition *network, std::map<std::string, W
 	return ew1->getOutput(0);
 }
 
-
 ITensor* RRDB(INetworkDefinition *network, std::map<std::string, Weights>& weightMap, ITensor* x, std::string lname)
 {
 	ITensor* out = residualDenseBlock(network, weightMap, x, lname + ".rdb1");
@@ -253,7 +252,24 @@ void createEngine(unsigned int maxBatchSize, IBuilder* builder, IBuilderConfig*
 
 	// Build engine
 	builder->setMaxBatchSize(maxBatchSize);
-	config->setMaxWorkspaceSize(1 << 20);
+	//config->setMaxWorkspaceSize(1 << 22);
+	config->setMaxWorkspaceSize(28 * (1 << 23));  // 28MB
+
+	if (precision_mode == 16) {
+		std::cout << "==== precision f16 ====" << std::endl << std::endl;
+		config->setFlag(BuilderFlag::kFP16);
+	}
+	else if (precision_mode == 8) {
+		//std::cout << "==== precision int8 ====" << std::endl << std::endl;
+		//std::cout << "Your platform support int8: " << builder->platformHasFastInt8() << std::endl;
+		//assert(builder->platformHasFastInt8());
+		//config->setFlag(BuilderFlag::kINT8);
+		//Int8EntropyCalibrator2 *calibrator = new Int8EntropyCalibrator2(maxBatchSize, INPUT_W, INPUT_H, 0, "../data_calib/", "../Int8_calib_table/detr_int8_calib.table", INPUT_BLOB_NAME);
+		//config->setInt8Calibrator(calibrator);
+	}
+	else {
+		std::cout << "==== precision f32 ====" << std::endl << std::endl;
+	}
 
 	std::cout << "Building engine, please wait for a while..." << std::endl;
 	IHostMemory* engine = builder->buildSerializedNetwork(*network, *config);
@@ -285,7 +301,7 @@ int main()
 	char engineFileName[] = "real-esrgan";
 
 	char engine_file_path[256];
-	sprintf(engine_file_path, "../Engine/%s.engine", engineFileName);
+	sprintf(engine_file_path, "../Engine/%s_%d.engine", engineFileName, precision_mode);
 
 	// 1) engine file 만들기 
 	// 강제 만들기 true면 무조건 다시 만들기
@@ -359,7 +375,7 @@ int main()
 	std::cout << "===== input load done =====" << std::endl << std::endl;
 
 	uint64_t dur_time = 0;
-	uint64_t iter_count = 1;
+	uint64_t iter_count = 10;
 
 	// CUDA 스트림 생성
 	cudaStream_t stream;