This page describes the tutorial of yolo-v4, up to the inference of the famous dog.ipg. os uses windows. Also, because I am not good at English, I may be taking unnecessary steps. If you are interested, please go to https://github.com/AlexeyAB/darknet and go by yourself. dog.jpg　↓↓↓

at first

It is assumed that basic things such as opencv and visual studio are done. If you haven't done it, check it out, the procedure is the same as up to v3.

Clone from git

Clone by typing the following command at the location where you want to build the darknet.

git clone https://github.com/AlexeyAB/darknet

do cmake

Go to the build in the darknet folder you cloned earlier and do cmake. The code below.

cd build
cmake ..

If cmake doesn't recognize it as a command, start by doing cmake. cmake is easy, so do your best to build it.

Build

There is a second darknet under the build folder, so it's in it

darknet.sln
yolo_console_dll.sln
yolo_cpp_dll.sln

We will build these three using visual studio. I think there is a file with no_gpu, but I haven't touched it so I don't know what will happen. As you can guess from the name, it will switch to cpu (-_-;) I will do it with gpu, so I will do it as gpu.

How to build each .sln

Take darknet.sln as an example. You can do the other two in the same way. If you open darknet.sln with vs, you will see a screen like this. 無題_LI.jpg procedure

Right-click on darknet in Solution Explorer in the upper right
Select your own cuda from source control
Right click again and select the bottom property
Open Configuration Manager in the upper right and change Active Solutions and Platforms to release and x64
Return to the property page and add the path with opencv and cuda include files to the include directory from c / c ++ (cuda include files are windows → program files → nvidia gpu → There is an include file in each version. opencv is self-help Please do your best)
After adding the include path, open the linker (on the left). There is an additional library directory in it, so add the opencv, cuda lib file path here
Press Apply
Right-click on darknet and press build (if you don't see anything here, press rebuild a few times and it will work).

After doing the above, if darknet.exe is created under x64, it is successful.

darknet.sln→darknet.exe
yolo_console_dll.sln→yolo_console_dll.exe
yolo_cpp_dll.sln→yolo_cpp_dll.dll

The above is made. At this point, it's almost over.

Try to infer

Execute the following command in the x64 folder.

python darknet.py

This command infers the default dog.jpg. result You will get results like this. I haven't made a correspondence table, so it's only bbox. If you want to do another image, you can do it by editing darknet.py.

After this

I want to let you learn, and I want to do this as well, so I will write if the number of articles does not increase when the learning is completed.

Impressions

It was difficult because there were few articles in Japanese (; ^ _ ^ A) I want to be able to study properly. I'm not good at English, so it's barely made, so the credibility is tattered (laughs), but if you are interested, please go by yourself from github. I feel like I'm doing a lot of things I don't need. I'm sorry it's hard to read m (_ _ "m)

Run yolov4 "for the time being" on windows