MNIST LeNet
C++ digit detection with opencv and dlib.
|
Detect Handwritten Digits using a LeNet trained with MNIST. This project comes with an simple HTML5 PWA which allows taking pictures and displaying the result.
(All of them are optional)
The path specified by the –mnist argument needs to contain the mnist dataset which can be obtained from here: http://yann.lecun.com/exdb/mnist/
On first startup a LeNet will be trained and stored within the path specified by –mnist. This network will be used for all predictions.
A pre trained network is provided (mnist/mnist_network.dat)
By default the PWA will only work with an valid SSL certificate (security restriction of most browsers).
To use the client without SSL certificate in chrome/chromium you need to enable the "Insecure origins treated as secure" setting for your host. To do so following steps need to be performed (this works on android aswell):
The PWA can then be accessed via the browser (on localhost via http://localhost:8808)
All provided executables are completely statically linked and thus should be able to run on any device running Linux/Windows.
Because the executable is completely statically linked, it can be executed on rooted android phones. Download a package matching your CPU architecture and copy it to '/data' using a terminal emulator or adb.
2020, Daniel Giritzer