Insights with building custom object detection models for large machine learning projects

Image annotation for building Visual AI. [Source]

Over time, we at Ximilar worked on many challenging projects incorporating object detection. Training and deployment of the model alone is nowadays possible in a matter of minutes. You can simply upload your data and click a few buttons to train and deploy your model as an API. Nowadays, the increasingly challenging and difficult part of detection models training is data, because:

  • All the items you want AI to detect, need to be properly annotated first, meaning they are somehow marked with labels and bounding boxes.
  • Annotation of data is…


Simple tutorial for detecting microcontrollers on data from Kaggle competition

Photo by Jonas Svidras from Pexels

Standart libraries as TensorFlow or PyTorch don’t provide any simple way to train your custom Object Detection models. Most of the time you need to install a big library as Detectron 2 or Tensorflow Object Detection API. Don’t get me wrong, these libraries are powerful but often requires a lot of inspecting, tuning, and playing with data. That is why I would like to show you a much simpler and effective way on how to train your detectors by calling Rest API and a bit of clicking.

There are advantages to this approach: you don’t need to be a machine…


AI will change all possible fields whether it is physics, law, or retail and one should be prepared for what is to come…

Photo by mentatdgt from Pexels

Half of this crazy year is behind us and summer is here. Over the years, we machine learning engineers at Ximilar have gathered a lot of interesting ML/AI material from which we draw. I have chosen the best ones from podcasts to online courses that I recommend to listen to, read, and check. Some of them are introductory, others more advanced. However, all of them are high-quality ones made by the best people in the field and they are worth checking. If you are interested in the current progress of AI or you are just curious about what will be…

Data Science in the Real World

Why every fashion site should embrace Machine Learning, how to work with taxonomy and data and how you can build an image recognition system with Flows

Photo by Thomas Serer on Unsplash

In this post, I will show you how you can build your Fashion recognition system. I will focus on footwear however this approach can be applied also on other Apparel. This approach is suitable not only in the Fashion domain. The great news is that this tutorial is code-free!

Why Machine Learning and Fashion?

There are several reasons: better user experience, increasing click-rate, engagement, revenue … Here is the big research what is the most important features in fashion retail that should every shop have:

As you can see, some of the features are purely based on Machine Learning…

The game-changing feature that makes deep learning more accessible. Solving computer vision problem of Screw & Nut recognition with hierarchical classification.

Photo by from Pexels

Nowadays, there are a lot of companies that are doing AI/Machine Learning. For example, many blue-chip companies are integrating their own machine learning solutions to their cloud. Using their cloud is becoming more and more complex and people started to become XYZ cloud service CERTIFIED specialists in order to use it.

“Everything should be made as simple as possible, but no simpler.”

Albert Einstein

Sometimes you or your team need to build accurate and more complex machine learning systems. But this is very hard when using AWS, Azure or AutoML consoles. The cloud solutions from big players have so many…

Making your networks to learn faster in TensorFlow 2+

source: Pexels

There are still a lot of models that use Batch Normalization layers. If you would like to do transfer learning on such models, you can have a lot of problems. It was much easier to finetune AlexNet or VGG as they do not contain batch norm layers. Many developers are curious why the more modern CNN architecture is not able to perform as well as the older one.

I ran to the problem of batch norm layers several times. I thought that something is wrong with the optimization of the model. Then I found the article by Datumbox about the…

A technique to make CNN models less prone to errors and more accessible for developers


One of the most common mistakes which are novice machine learning practitioners doing is forget to normalize input images. No wonder! Every model requires different input normalization when you are doing Transfer Learning. VGG, for example, requires to subtract this vector [123.68, 116.779, 103.939] from the RGB image. MobileNetV2 requires inputs from interval <-1,1>. PyTorch models often use different normalization methods. This tip can save you a lot of time and bugs in the future. Instead of doing normalization before inputting images to the model, you can simply add this layer inside your model (computation graph).

With old TensorFlow 1…

Building a visual quality control system to check your products on a few clicks with Convolutional Neural Networks and Ximilar platform.

Photo by FWStudio

Building visual inspection system is the common problem in lot of factories and Machine Learning approach is scalable solution. Not only your production process can be automated, it can also create more high quality products …

That is why this blog post will be focused on recognizing defects on the images with Ximilar platform. We are going to show how easy is to build an image quality control model. The visual quality control can be built on other types of images like:

  • any products/objects as screws, automotive parts, toys…

Michal Lukac

Co-Founder & Machine Learning developer at Ximilar. Brazillian Jiu-Jitsu practitioner, book lover, geek, investing in stocks.

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store