add yolov3-tiny

Author: wang-xinyu

yolov3-tiny/CMakeLists.txt

@@ -0,0 +1,42 @@
+cmake_minimum_required(VERSION 2.6)
+
+project(yolov3-tiny)
+
+add_definitions(-std=c++11)
+
+option(CUDA_USE_STATIC_CUDA_RUNTIME OFF)
+set(CMAKE_CXX_STANDARD 11)
+set(CMAKE_BUILD_TYPE Debug)
+
+find_package(CUDA REQUIRED)
+
+set(CUDA_NVCC_PLAGS ${CUDA_NVCC_PLAGS};-std=c++11;-g;-G;-gencode;arch=compute_30;code=sm_30)
+
+include_directories(${PROJECT_SOURCE_DIR}/include)
+if (CMAKE_SYSTEM_PROCESSOR MATCHES "aarch64")
+    message("embed_platform on")
+    include_directories(/usr/local/cuda/targets/aarch64-linux/include)
+    link_directories(/usr/local/cuda/targets/aarch64-linux/lib)
+else()
+    message("embed_platform off")
+    include_directories(/usr/local/cuda/include)
+    link_directories(/usr/local/cuda/lib64)
+endif()
+
+
+set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11 -Wall -Ofast -Wfatal-errors -D_MWAITXINTRIN_H_INCLUDED")
+
+#cuda_add_library(leaky ${PROJECT_SOURCE_DIR}/leaky.cu)
+cuda_add_library(yololayer SHARED ${PROJECT_SOURCE_DIR}/yololayer.cu)
+
+find_package(OpenCV)
+include_directories(OpenCV_INCLUDE_DIRS)
+
+add_executable(yolov3-tiny ${PROJECT_SOURCE_DIR}/yolov3-tiny.cpp)
+target_link_libraries(yolov3-tiny nvinfer)
+target_link_libraries(yolov3-tiny cudart)
+target_link_libraries(yolov3-tiny yololayer)
+target_link_libraries(yolov3-tiny ${OpenCV_LIBS})
+
+add_definitions(-O2 -pthread)
+

yolov3-tiny/README.md

@@ -0,0 +1,49 @@
+# yolov3-tiny
+
+The Pytorch implementation is [ultralytics/yolov3](https://github.com/ultralytics/yolov3).
+
+## Excute:
+
+```
+1. generate yolov3-tiny.wts from pytorch implementation with yolov3-tiny.cfg and yolov3-tiny.weights
+
+git clone https://github.com/ultralytics/yolov3.git
+// download its weights 'yolov3-tiny.pt' or 'yolov3-tiny.weights'
+// put tensorrtx/yolov3-tiny/gen_wts.py into ultralytics/yolov3 and run
+python gen_wts.py yolov3-tiny.weights
+// a file 'yolov3-tiny.wts' will be generated.
+
+2. put yolov3-tiny.wts into tensorrtx/yolov3-tiny, build and run
+
+// go to tensorrtx/yolov3-tiny
+mkdir build
+cd build
+cmake ..
+make
+sudo ./yolov3-tiny -s             // serialize model to plan file i.e. 'yolov3-tiny.engine'
+sudo ./yolov3-tiny -d  ../../yolov3-spp/samples // deserialize plan file and run inference, the images in samples will be processed.
+
+3. check the images generated, as follows. _zidane.jpg and _bus.jpg
+```
+
+<p align="center">
+<img src="https://user-images.githubusercontent.com/15235574/78247927-4d9fac00-751e-11ea-8b1b-704a0aeb3fcf.jpg">
+</p>
+
+<p align="center">
+<img src="https://user-images.githubusercontent.com/15235574/78247970-60b27c00-751e-11ea-88df-41473fed4823.jpg">
+</p>
+
+## Config
+
+- Input shape defined in yololayer.h
+- Number of classes defined in yololayer.h
+- FP16/FP32 can be selected by the macro in yolov3-tiny.cpp
+- GPU id can be selected by the macro in yolov3-tiny.cpp
+- NMS thresh in yolov3-tiny.cpp
+- BBox confidence thresh in yolov3-tiny.cpp
+
+## More Information
+
+See the readme in [home page.](https://github.com/wang-xinyu/tensorrtx)
+

yolov3-tiny/gen_wts.py

@@ -0,0 +1,24 @@
+import struct
+import sys
+from models import *
+from utils.utils import *
+
+model = Darknet('cfg/yolov3-tiny.cfg', (608, 608))
+weights = sys.argv[1]
+dev = '0'
+if weights.endswith('.pt'):  # pytorch format
+    model.load_state_dict(torch.load(weights, map_location=device)['model'])
+else:  # darknet format
+    load_darknet_weights(model, weights)
+model = model.eval()
+
+f = open('yolov3-tiny.wts', 'w')
+f.write('{}\n'.format(len(model.state_dict().keys())))
+for k, v in model.state_dict().items():
+    vr = v.reshape(-1).cpu().numpy()
+    f.write('{} {} '.format(k, len(vr)))
+    for vv in vr:
+        f.write(' ')
+        f.write(struct.pack('>f',float(vv)).hex())
+    f.write('\n')
+