Still no support for Windows?
jaroslavhuss opened this issue · comments
Hello!,
when trying to instal darknet, I am getting this error:
`C:\Users\hussj\Desktop\MLTest>npm i darknet
npm WARN deprecated request@2.88.2: request has been deprecated, see request/request#3142
npm WARN deprecated har-validator@5.1.5: this library is no longer supported
darknet@3.1.6 install C:\Users\hussj\Desktop\MLTest\node_modules\darknet
./install-script.sh && node-gyp rebuild
'.' is not recognized as an internal or external command,
operable program or batch file.
npm WARN MLTest@1.0.0 No description
npm WARN MLTest@1.0.0 No repository field.
npm ERR! code ELIFECYCLE
npm ERR! errno 1
npm ERR! darknet@3.1.6 install: ./install-script.sh && node-gyp rebuild
npm ERR! Exit status 1
npm ERR!
npm ERR! Failed at the darknet@3.1.6 install script.
npm ERR! This is probably not a problem with npm. There is likely additional logging output above.
npm ERR! A complete log of this run can be found in:
npm ERR! C:\Users\hussj\AppData\Roaming\npm-cache_logs\2021-09-07T18_09_13_373Z-debug.log`
It seems it does not want run node-gyp command... Is it because I am on W10?
Thank you!
Hey, no there's still no native support for Windows. There's an issue (#25) for supporting Windows but since I run Linux I haven't gotten around to it.
As far as I'm aware darknet.js should run fine in the WSL so that could be an option in the meantime?
@bennetthardwick - it was great advice to use WSL. I succeeded in the setup up of your Darknet.js, unfortunately, it takes 12-15 seconds to evaluate the image. I am not sure what the problem is - maybe you could give me another advice?
I have created this repo. I will try to ask some more people what could be the issue or what would be the best approach for the project I am having fun with.
Thank you again Bennet for your advice and really good work!
Jarda
Hey, @jaroslavhuss it looks you're using a config that has 106 layers. The examples in this repo all use the "yolov3-tiny" config which is built for speed - it only has 24 layers.
You can see in this graph from pjreddie's website that using yolov3 they get about 20 - 40 fps but using yolov3-tiny they get 220 fps.
You should consider retraining using the yolov3-tiny config, or even the yolov4-tiny config (although I've not tested it before)
Also if you have a CUDA compatible GPU you should look into enabling GPU detections - it dramatically improves the speed of detections. Here's some info on CUDA for WSL as well as how to enable it for darknet.js:
Best of luck! Let me know how you go.
Dear Bennett,
your answer helped a lot! I actually tried your examples, and with WSL2 without CUDA i got myself in about 1 second. Which is perfect!
Anyway, I think I am facing one last issue - and that is the config file and google collab.
My Google Collab sheet can be found here - especially point number 3) Configure Darknet network for training YOLO V3 - I am using pjreddies cfg file, but the problem is that I am using just one class called "monster".
Basically even after 1,5k iterations, the algorithm is not able to find any "monster" and I believe I still have some errors in the aforementioned file.
Please, if you find some time to look at that google collab, I would be super grateful.
Ps.:
- Link to new trained data.
- Link to the labeled images
- Link to the cfg
- Link to the google collab sheet
Thank you in advance!
Hey @jaroslavhuss, nothing wrong really stands out to me...
Have you tried training without changing the config too much? Perhaps try only uncommenting these lines: https://github.com/AlexeyAB/darknet/blob/9ff8653d999c8a22bc8f1ff4f4a8a3cc5b63d255/cfg/yolov3-tiny.cfg#L2-L7
Other than that I'm not really sure, unfortunately. The last time I successfully trained my own model was three years ago and I remember there was a lot of trial and error.
@bennetthardwick it works! You helped a lot! All I need to do is to speed WSL2 with GPU and it will be fast :-). Sir, you are freaking awesome! If you ever pay a visit to the Czech Republic, please, let me know - I will be your guide :-). Many and many thanks from a really happy person.
Awesome, that's great to hear! I'm sure I'll take you up on that offer one day 😆