::: tiny-dnn based C++ deep learning applications :::

After spending more than a year working on a from scratch deep learning C++ stack (neurocl), I decided to switch my experiments to use a more established C++ framework, that is tiny-dnn.

tiny-brain additional image processing is implemented in header only tinymage class, which is based on std::vector private inheritance, and allows some basic image processing (thresholding, resizing, rotating, cropping, computing row/column sum images...). Some more task specific processing is implemented in header only classes tinydigit and tinysign, respectively dedicated to digits recognition and digits sign extraction (from a rich scene).

tiny-brain is C++14 compliant.

Linux

Cloning

tiny-brain includes tiny-dnn as a git submodule, therefore use the following command line for recursive cloning:

git clone --recursive https://github.com/blackccpie/tiny-brain.git

Demos

Full-webassemby mnist digit recognition demo can be viewed online here:

Digits localization and recognition on white background

Digits sign localization, extraction, warping and recognition (in progress...3/4 correct guesses)

Status

Development is on hiatus while facing some functional issues related to #857. Hope this one will be fixed soon.