Tensorflow implementation on Windows UWP x86
We are always trying to make our customers happy of course – but this time it wasn’t that easy. Tensorflow is only quite recently supported on Windows in general.
So how should we manage to run our software on 32 bit Windows systems through Tensorflow when their build system is just designed for 64 bit?
Usually, most of the time it makes sense to go with 64 bit and not 32 bit, especially when it comes to large computations such as neural networks – but for this task we need to – and so our developers did it by themselves and they succeeded!
Last week, we proudly announced that we built the worldwide first (as far as we know ? ) Tensorflow implementation on Windows UWP with 32 bit architecture.
Usually the powerful Tensorflow framework, which is a C++ library, is called via python wrappers. So what our developers did was to consume Tensorflow directly in our C++ library without python. Our library is compiled for x86, so we are able to support desktop devices such as the Microsoft Surface as well as the Microsoft HoloLens, which only runs on 32 bit.
What is Tensorflow and what makes it so special
Tensorflow is an open-sourced software developed by the Google Brain team. Since 2015 it’s possible for everyone to use Tensorflow for free. With this software you are able to construct neural networks, which allows you to train data fast and easy. Tensorflow is usually used as a Python package, but can also be built from source and be consumed in C++ code. So without Tensorflow thousands of companies and projects wouldn’t have the ability to make as big steps as they do. It’s easier than ever to use machine learning and computer vision within this software and apply it to real world use cases. Here at Anyline Tensorflow is an important part of our development and we are curious about the future of deep learning software.
Why is the combination of Anyline & Tensorflow Kickass
By using Tensorflow for our OCR SDK it guarantees a higher performance. Our development department trains neural networks with a great data volume and is able to use this trained network for our software. So with Tensorflow we find a perfect tool, which provides an awesome high performance and accurate way of doing (offline) OCR in real time. Through its architecture it can be trained and optimized consistently and with each update the Anyline SDK will be more powerful. If there are some barriers (Tensorflow wasn’t running on Windows UWP x86 32bit), we don’t back off but break down those barriers to get the best OCR SDK on the market.
QUESTIONS? LET US KNOW!
Sign up for our Newsletter!
Be the first to receive the latest news, release notes and updates!