alexeyab / yolo-windows Goto Github PK

View Code? Open in Web Editor NEW

This project forked from frischzenger/yolo-windows

70.0 8.0 58.0 1.59 MB

a yolo windows version(for object detection)

Shell 0.12% Python 0.53% Cuda 8.22% C 90.56% Makefile 0.23% Batchfile 0.10% C++ 0.25%

yolo-windows's Introduction

Yolo-Windows v1

You can get new more accurate and faster Yolo v2 Windows by the link: https://github.com/AlexeyAB/darknet

"You Only Look Once: Unified, Real-Time Object Detection"

Here is a Yolo v1 windows version (for object detection)

Contributtors: https://github.com/pjreddie/darknet/graphs/contributors

This repository is forked from: https://github.com/frischzenger/yolo-windows Which is forked from Linux-version: https://github.com/pjreddie/darknet

More details: http://pjreddie.com/darknet/yolov1/

Requires:

MS Visual Studio 2013 (v120)
CUDA 7.5 for Windows x64: https://developer.nvidia.com/cuda-75-downloads-archive
OpenCV 2.4.9: https://sourceforge.net/projects/opencvlibrary/files/opencv-win/2.4.9/opencv-2.4.9.exe/download
- To compile without OpenCV - remove define OPENCV from: Visual Studio->Project->Properties->C/C++->Preprocessor
- To compile with different OpenCV version - change in file yolo_kernels.cu each string #pragma comment(lib, "opencv_core249.lib") from 249 to required version.
- With OpenCV will show image or video detection in window and store result to test_dnn_out.avi

Pre-trained models for different cfg-files can be downloaded from (smaller -> faster & lower quality):

yolo.cfg (1 GB) - require 4 GB GPU-RAM: https://drive.google.com/open?id=0BwRgzHpNbsWBLU04Znc4U1BMZ0k
yolo-small.cfg (359 MB) - require 2 GB GPU-RAM: https://drive.google.com/open?id=0BwRgzHpNbsWBQzlHeVpZRnhlY2M
yolo-tiny.cfg (172 MB) - require 1 GB GPU-RAM: https://drive.google.com/open?id=0BwRgzHpNbsWBQlMxcXNBMWtkaFk

Examples of results:

Others: https://www.youtube.com/channel/UC7ev3hNVkx4DzZ3LO19oebg

Example of usage in cmd-files:

darknet.cmd - initialization with 1 GB model yolo.weights & yolo.cfg and get command prompt: enter the jpg-image-filename
darknet_demo.cmd - initialization with 1 GB model yolo.weights & yolo.cfg and play your video file which you must rename to: test.mp4
darknet_demo_small.cmd - initialization with 359 MB small model yolo-small.weights & yolo-small.cfg and play your video file which you must rename to: test.mp4
darknet_net_cam.cmd - initialization with 1 GB model, play video from network video-camera mjpeg-stream and store result to: test_dnn_out.avi

For using network video-camera mjpeg-stream with any Android smartphone:

Download for Android phone mjpeg-stream soft: IP Webcam / Smart WebCam

Smart WebCam - preferably: https://play.google.com/store/apps/details?id=com.acontech.android.SmartWebCam IP Webcam: https://play.google.com/store/apps/details?id=com.pas.webcam

Connect your Android phone to computer by WiFi (through a WiFi-router) or USB
Start Smart WebCam on your phone
Replace the address below, on shown in the phone application (Smart WebCam) and launch:

darknet.exe yolo demo yolo.cfg yolo.weights http://192.168.0.80:8080/video?dummy=param.mjpg -i 0

Windows (Visual Studio) Support Now yolo supports windows with Visual Studio. Just change the include directories / library directories to your own ones, and compile the codes with X64 Release mode. You may need pthread-windows while compiling and linking

How to train yourself:

To training for your 20 objects download pre-trained convolutional layers:

darknet.conv.weights: https://drive.google.com/open?id=0BwRgzHpNbsWBUXdmdXdaa1dDU2s
extraction.conv.weights: https://drive.google.com/open?id=0BwRgzHpNbsWBaHR1NjZ2T3FKT1E

Then use the following command lines:

For tiny-model: darknet yolo train yolo-tiny.cfg darknet.conv.weights
For small-model: darknet yolo train yolo-small.cfg extraction.conv.weights
For full-model: darknet yolo train yolo.cfg extraction.conv.weights

To train for a different number of objects (not 20) or for different number of sides (not 7), you should change your .cfg-file 3 lines:

output= (5 x 2 + C) x S x S
classes= C
side= S

Where:

C = number of objects (by default 20)
S = number of sides (by default 7)

More details: http://pjreddie.com/darknet/yolov1/

How to compile:

If you have CUDA 7.5, OpenCV 2.4.9 (C:\opencv_2.4.9) and MSVS 2013 (or 2015) then start MSVS, open yolo-windows\build\darknet\darknet.sln and do the: Build -> Build darknet
If you have other version of CUDA (not 7.5) then open yolo-windows\build\darknet\darknet\darknet.vcxproj by using Notepad, find 2 places with "CUDA 7.5" and change it to your CUDA-version, then do step 1
If you have other version of OpenCV 2.4.x (not 2.4.9) then you should change pathes after \darknet.sln is opened

3.1 (right click on project) -> properties -> C/C++ -> General -> Additional Include Directories

3.2 (right click on project) -> properties -> Linker -> General -> Additional Library Directories

If you have other version of OpenCV 3.x (not 2.4.x) then you should change many places in code by yourself.

How to compile (custom):

Also, you can to create your own darknet.sln & darknet.vcxproj, this example for CUDA 7.5 and OpenCV 2.4.9

Then add to your created project:

(right click on project) -> properties -> C/C++ -> General -> Additional Include Directories, put here:

C:\opencv_2.4.9\opencv\build\include;..\..\..\3rdparty\include;%(AdditionalIncludeDirectories);$(CudaToolkitIncludeDir)

right click on project -> Build dependecies -> Build Customizations -> set check on CUDA 7.5 or what version you have - for example as here: http://devblogs.nvidia.com/parallelforall/wp-content/uploads/2015/01/VS2013-R-5.jpg
add to project all .c & .cu files from yolo-windows\src
(right click on project) -> properties -> Linker -> General -> Additional Library Directories, put here:

C:\opencv_2.4.9\opencv\build\x64\vc12\lib;$(CUDA_PATH)lib\$(PlatformName);%(AdditionalLibraryDirectories)

(right click on project) -> properties -> Linker -> Input -> Additional dependecies, put here:

..\..\..\3rdparty\lib\x64\pthreadVC2.lib;cublas.lib;curand.lib;cudart.lib;%(AdditionalDependencies)

(right click on project) -> properties -> C/C++ -> Preprocessor -> Preprocessor Definitions

OPENCV;_CRT_SECURE_NO_WARNINGS;GPU;WIN32;NDEBUG;_CONSOLE;_LIB;%(PreprocessorDefinitions)

compile to .exe (X64 & Release) and put .dll`s near with .exe:

pthreadVC2.dll, pthreadGC2.dll from yolo-windows\3rdparty\dll\x64

cusolver64_75.dll, curand64_75.dll, cudart64_75.dll, cublas64_75.dll - 75 for CUDA 7.5 or your version, from C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v7.5\bin

yolo-windows's People

Contributors

Stargazers

Watchers

yolo-windows's Issues

why a lot of classes has changed in windows version like removing CUDNN ?

I noticed that a lot of classes has changed ?
the most important changed that the code now dosen't support CUDNN,
that's why the training and testing is slow?

Compile errors with yolo2_light

I have some trouble with yolo2_light,

in README.md, it says
To compile for CPU just do make on Linux

when I do that it turns out

find . -name "*.sh" -exec chmod +x {} \;
gcc  -Wall -Wfatal-errors -Ofast obj/main.o obj/additionally.o obj/box.o obj/yolov2_forward_network.o -o bin/darknet -lm -pthread 
obj/main.o: In function `test_detector_cpu':
main.c:(.text+0x4e0): undefined reference to `network_predict_quantized'
collect2: error: ld returned 1 exit status
make: *** [bin/darknet] Error 1

I wonder if I should do sth. to Makefile or there're any other solutions?

Memory Leak

Hi, I compiled YOLO on windows and it works well. I am using it for object detection using webcam in real-time.

But, the memory usage by Darknet (YOLO) in Windows slowly keeps on increasing and about after 10 minutes my computer runs out of memory (16 GB RAM). I was wondering if you faced similar issue.

How to make yolo cpp dll thread safe

Hello

I'm using yolo_cpp_dll_no_gpu.dll in a secondary c++ project. If I have one thread using this Dll every thing is OK, but how can I use darknet functions in multiple threads?

I'm using c++ and not much familiar with C.
I tried to make an Interface c++ object, in which I can use the dll in multiple threads but, only with mutexes and lockes, but of course the performance will be compromised.

Thanks.

How to test image?

when initialization with 1 GB model yolo.weights & yolo.cfg

it shows as follows:

Cannot load image "data/labels/aeroplane.png"
STB Reason: can't fopen

problem when i compile darknet

when i compile the darknet.it comes out many errors such as
错误 114 error : this declaration may not have extern "C" linkage C:\Program Files (x86)\Microsoft Visual Studio 12.0\VC\include\limits 78 1 darknet2
in limits iosfwd and type_traits
and something wrong in avgpool_layer

I didn't get any detcted boxes after training new data

I trained new data using "train_yolo" method,
I changed the 'CLASSNUM' in yolo.c and 'classes' in .cfg file
Then changed output in .cfg in [connected] based on this equation (5 * 2 + C) * S * S

after training 600 batches I tested the "yolo_600.weight" file and It didn't detect anything

I used same .cfg file of training for testing..

how many batches I should train until I start get any result?
is there any another steps I should made?
I even put "-thresh 0" and didn't get any box??
there is an article said that I should change CLS_NUM in yolo_kernels.cu as well .. but I didn't see this parameter?

Unreasonable high score for the classes

I have already used your yolo.cfg and download the corresponding yolo.weights but get some scores as below.

Do you know how to solve it?

Do you know how to run ssd on windows?

Thank you for your project. It runs good on Windows. I’m working on ssd (Single Shot MultiBox Detector)now. Accuracy rate of ssd is as same as that of faster Rcnn while its speed is as quick as yolo’s. But some problems occur when migrating it to Windows(boost lib problems ). Have you thought about migrating ssd to Window?
This is home page of the ssd project: https://github.com/weiliu89/caffe/tree/ssd
System VOC2007 mAP FPS (Titan X)
Faster R-CNN (VGG16) 73.2 7 7
YOLO 63.4 45
SSD300 (VGG16) 72.1 58
SSD500 (VGG16) 75.1 23

How can I use it without GPU

hello guys,
I know the GPU is very very good at YOLO, but.... I want to know how compiler it without GPU. When I delete GPU in precompiler, it will show a lot of error such as "identifier BLOCK is undefined" and "check_error" is undefined and so on.

OpenCV 3.3 Compile Errors

Using CUDA 8.0 and Visual Studio 2015, OpenCV 3.3.0:

Anyone else having this error when compiling?

cannot open source file 'opencv2/highgui/highgui_c.h'
cannot open source file 'opencv2/imgproc/imgproc_c.h'

I've set the additional directories in the project settings and the linker.

Installation help

I'm sorry, I'm relatively new to Visual Studio and am unsure of what steps I should take to compile. Is there a step by step guide? If not, currently I have CUDA 6.5 x64 installed (as well as 7.5) and OpenCV 2.4.9, and I'm trying to load the yolo-windows-master\build\darknet\darknet.sln with VS 2013, but to no luck.

C:\Users(username)\Downloads\yolo-windows-master (vs2013, Cuda 6.5, Git AlexeyAB)\yolo-windows-master\build\darknet\darknet\darknet.vcxproj : error : Unable to read the project file "darknet.vcxproj".

C:\Users(username)\Downloads\yolo-windows-master (vs2013, Cuda 6.5, Git AlexeyAB)\yolo-windows-master\build\darknet\darknet\darknet.vcxproj(55,5): The imported project "C:\Program Files (x86)\MSBuild\Microsoft.Cpp\v4.0\V120\BuildCustomizations\CUDA 6.5.props" was not found. Confirm that the path in the declaration is correct, and that the file exists on disk.

Any help is much appreciated. Thank you for your time.

EDIT: I wanted to add that I've installed YOLO on Ubuntu, so I have a (very vague) idea of how it should work. Main problem with that installation is, I could get my GPU to function within Virtual Box, and because I need web cam real-time detection in my work, I'm trying to install YOLO on windows instead.

Is it possible to perform detection and test on part of each image?

How can I select a specific region of image (manually) to perform all the operations on that region and ignore the rest of the image?