Add files via upload

Author: KelvinCPChiu

DeepStream/nvdsinfer_custom_impl_Yolo/kernels.cu

@@ -0,0 +1,77 @@
+/*
+ * Copyright (c) 2018-2019 NVIDIA Corporation.  All rights reserved.
+ *
+ * NVIDIA Corporation and its licensors retain all intellectual property
+ * and proprietary rights in and to this software, related documentation
+ * and any modifications thereto.  Any use, reproduction, disclosure or
+ * distribution of this software and related documentation without an express
+ * license agreement from NVIDIA Corporation is strictly prohibited.
+ *
+ */
+
+#include <cuda.h>
+#include <cuda_runtime.h>
+#include <stdint.h>
+#include <stdio.h>
+#include <string.h>
+
+inline __device__ float sigmoidGPU(const float& x) { return 1.0f / (1.0f + __expf(-x)); }
+
+__global__ void gpuYoloLayerV3(const float* input, float* output, const uint gridSize, const uint numOutputClasses,
+                               const uint numBBoxes)
+{
+    uint x_id = blockIdx.x * blockDim.x + threadIdx.x;
+    uint y_id = blockIdx.y * blockDim.y + threadIdx.y;
+    uint z_id = blockIdx.z * blockDim.z + threadIdx.z;
+
+    if ((x_id >= gridSize) || (y_id >= gridSize) || (z_id >= numBBoxes))
+    {
+        return;
+    }
+
+    const int numGridCells = gridSize * gridSize;
+    const int bbindex = y_id * gridSize + x_id;
+
+    output[bbindex + numGridCells * (z_id * (5 + numOutputClasses) + 0)]
+        = sigmoidGPU(input[bbindex + numGridCells * (z_id * (5 + numOutputClasses) + 0)]);
+
+    output[bbindex + numGridCells * (z_id * (5 + numOutputClasses) + 1)]
+        = sigmoidGPU(input[bbindex + numGridCells * (z_id * (5 + numOutputClasses) + 1)]);
+
+    output[bbindex + numGridCells * (z_id * (5 + numOutputClasses) + 2)]
+        = __expf(input[bbindex + numGridCells * (z_id * (5 + numOutputClasses) + 2)]);
+
+    output[bbindex + numGridCells * (z_id * (5 + numOutputClasses) + 3)]
+        = __expf(input[bbindex + numGridCells * (z_id * (5 + numOutputClasses) + 3)]);
+
+    output[bbindex + numGridCells * (z_id * (5 + numOutputClasses) + 4)]
+        = sigmoidGPU(input[bbindex + numGridCells * (z_id * (5 + numOutputClasses) + 4)]);
+
+    for (uint i = 0; i < numOutputClasses; ++i)
+    {
+        output[bbindex + numGridCells * (z_id * (5 + numOutputClasses) + (5 + i))]
+            = sigmoidGPU(input[bbindex + numGridCells * (z_id * (5 + numOutputClasses) + (5 + i))]);
+    }
+}
+
+cudaError_t cudaYoloLayerV3(const void* input, void* output, const uint& batchSize, const uint& gridSize,
+                            const uint& numOutputClasses, const uint& numBBoxes,
+                            uint64_t outputSize, cudaStream_t stream);
+
+cudaError_t cudaYoloLayerV3(const void* input, void* output, const uint& batchSize, const uint& gridSize,
+                            const uint& numOutputClasses, const uint& numBBoxes,
+                            uint64_t outputSize, cudaStream_t stream)
+{
+    dim3 threads_per_block(16, 16, 4);
+    dim3 number_of_blocks((gridSize / threads_per_block.x) + 1,
+                          (gridSize / threads_per_block.y) + 1,
+                          (numBBoxes / threads_per_block.z) + 1);
+    for (unsigned int batch = 0; batch < batchSize; ++batch)
+    {
+        gpuYoloLayerV3<<<number_of_blocks, threads_per_block, 0, stream>>>(
+            reinterpret_cast<const float*>(input) + (batch * outputSize),
+            reinterpret_cast<float*>(output) + (batch * outputSize), gridSize, numOutputClasses,
+            numBBoxes);
+    }
+    return cudaGetLastError();
+}

DeepStream/nvdsinfer_custom_impl_Yolo/nvdsinfer_yolo_engine.cpp

@@ -0,0 +1,118 @@
+/*
+ * Copyright (c) 2019-2020, NVIDIA CORPORATION. All rights reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ */
+
+#include "nvdsinfer_custom_impl.h"
+#include "nvdsinfer_context.h"
+#include "yoloPlugins.h"
+#include "yolo.h"
+
+#include <algorithm>
+
+#define USE_CUDA_ENGINE_GET_API 1
+
+static bool getYoloNetworkInfo (NetworkInfo &networkInfo, const NvDsInferContextInitParams* initParams)
+{
+    std::string yoloCfg = initParams->customNetworkConfigFilePath;
+    std::string yoloType;
+
+    std::transform (yoloCfg.begin(), yoloCfg.end(), yoloCfg.begin(), [] (uint8_t c) {
+        return std::tolower (c);});
+
+    if (yoloCfg.find("yolov2") != std::string::npos) {
+        if (yoloCfg.find("yolov2-tiny") != std::string::npos)
+            yoloType = "yolov2-tiny";
+        else
+            yoloType = "yolov2";
+    } else if (yoloCfg.find("yolov3") != std::string::npos) {
+        if (yoloCfg.find("yolov3-tiny") != std::string::npos)
+            yoloType = "yolov3-tiny";
+        else
+            yoloType = "yolov3";
+    } else {
+        std::cerr << "Yolo type is not defined from config file name:"
+                  << yoloCfg << std::endl;
+        return false;
+    }
+
+    networkInfo.networkType     = yoloType;
+    networkInfo.configFilePath  = initParams->customNetworkConfigFilePath;
+    networkInfo.wtsFilePath     = initParams->modelFilePath;
+    networkInfo.deviceType      = (initParams->useDLA ? "kDLA" : "kGPU");
+    networkInfo.inputBlobName   = "data";
+
+    if (networkInfo.configFilePath.empty() ||
+        networkInfo.wtsFilePath.empty()) {
+        std::cerr << "Yolo config file or weights file is NOT specified."
+                  << std::endl;
+        return false;
+    }
+
+    if (!fileExists(networkInfo.configFilePath) ||
+        !fileExists(networkInfo.wtsFilePath)) {
+        std::cerr << "Yolo config file or weights file is NOT exist."
+                  << std::endl;
+        return false;
+    }
+
+    return true;
+}
+
+#if !USE_CUDA_ENGINE_GET_API
+IModelParser* NvDsInferCreateModelParser(
+    const NvDsInferContextInitParams* initParams) {
+    NetworkInfo networkInfo;
+    if (!getYoloNetworkInfo(networkInfo, initParams)) {
+      return nullptr;
+    }
+
+    return new Yolo(networkInfo);
+}
+#else
+extern "C"
+bool NvDsInferYoloCudaEngineGet(nvinfer1::IBuilder * const builder,
+        const NvDsInferContextInitParams * const initParams,
+        nvinfer1::DataType dataType,
+        nvinfer1::ICudaEngine *& cudaEngine);
+
+extern "C"
+bool NvDsInferYoloCudaEngineGet(nvinfer1::IBuilder * const builder,
+        const NvDsInferContextInitParams * const initParams,
+        nvinfer1::DataType dataType,
+        nvinfer1::ICudaEngine *& cudaEngine)
+{
+    NetworkInfo networkInfo;
+    if (!getYoloNetworkInfo(networkInfo, initParams)) {
+      return false;
+    }
+
+    Yolo yolo(networkInfo);
+    cudaEngine = yolo.createEngine (builder);
+    if (cudaEngine == nullptr)
+    {
+        std::cerr << "Failed to build cuda engine on "
+                  << networkInfo.configFilePath << std::endl;
+        return false;
+    }
+
+    return true;
+}
+#endif