stephanecharette / darkplate Goto Github PK

Hello Dear Mr. Stéphane Charette,

I tried the DarkPlate executable app with the config files and weghts you provided and it works! But if I replace them with the files and weights from this url (http://sergiomsilva.com/pubs/alpr-unconstrained/) it doesn't work.

App exits here:

bahman@Omen:~/projects/DarkPlate/build/src$ ./DarkPlate ../../media/m2.mp4
Looking for ./DarkPlate.cfg
Looking for ../DarkPlate.cfg
Looking for ../../DarkPlate.cfg
Looking for nn/DarkPlate.cfg
Looking for ../nn/DarkPlate.cfg
Looking for ../../nn/DarkPlate.cfg
Found neural network: ../../nn/DarkPlate.cfg

How can I adopt aforementioned pretrained models? Thanks in advance!

Hi Stephan,

Love the youtube videos and the projects you have created. I am new and working on a hobby project of mine. I installed all the pre requirements of DarkPlate but when I run sudo make I get the following:
cosme@yolo:~/src/darkplate/build$ sudo make
Consolidate compiler generated dependencies of target DarkPlate
[ 50%] Building CXX object src/CMakeFiles/DarkPlate.dir/main.cpp.o
/home/cosme/src/darkplate/src/main.cpp: In function ‘void process_plate(DarkHelp::NN&, cv::Mat&, cv::Mat&)’:
/home/cosme/src/darkplate/src/main.cpp:77:25: error: loop variable ‘prediction’ creates a copy from type ‘const DarkHelp::PredictionResult’ [-Werror=range-loop-construct]
77 | for (const auto prediction : results)
| ^~~~~~~~~~
/home/cosme/src/darkplate/src/main.cpp:77:25: note: use reference type to prevent copying
77 | for (const auto prediction : results)
| ^~~~~~~~~~
| &
cc1plus: all warnings being treated as errors
make[2]: *** [src/CMakeFiles/DarkPlate.dir/build.make:76: src/CMakeFiles/DarkPlate.dir/main.cpp.o] Error 1
make[1]: *** [CMakeFiles/Makefile2:98: src/CMakeFiles/DarkPlate.dir/all] Error 2
make: *** [Makefile:91: all] Error 2

Any ideas? My guess is user error :)

Thanks,
Cosme

Good day! Thank you very much for the tool, it is very useful and necessary! How can I run multiple video sources?

Or at least how can you run the code with the rtsp camera?

P.S Please make a donation button on the channel so that we can support you! Thanks again)

Hello, Stephane! Could you tell me how to train a model only on a gray channel? I do not need RGB, as many experiments have shown that special modes of compensation of light from the camera in the black and white image mode work much better than the color image. In this regard, it is necessary to test the model with only the gray channel. What do I need to change to get a single channel model? Thanks in advance!

Hi Stephane,

This is really good, just one issue. How can I reduce the input video size to make it run smoothly.
Right now its running the input video in full 1080p mode, hence slow.
Is there any way to pass the video in lower resolution to the inference model so as to make it realtime ?

Hi Stephan,

I copied my images directory in DarkPlate/nn directory , but while loading the images I can see only 4 classes :-

bicycle
car
truck
person

There is no "plate" or number/alphabet class at all. How to add these classes ?

Could you explain better the idea behind this projected? As far I am concerned, it looks like you :

1. Trained YOLO to recognize (Letters/Numbers + License Plate)
2. Parse video, detect each frame
3. Filter only detections inside 'License Plate detection bbox'
4. Read detections labels from left to right

Hey there, nice job first of all.

Im wondering will it work of any kind of license plates? I mean, different from countries

Hi Stephan,

Can I use your previous DarkPlate_best.weights while training new labelled images ?
I mean does DarkPlate support trasnfer learning ?

Dear Stephan:
I think you used the yolov4-tiny model for training. I want to use yolov4 model to train the license plate detection model. Would you please provide your labeled dataset? Thanks a lot!