Due to advances in the field of machine learning in recent years, any pattern in image data visible to the human eye can also be made visible to a machine. Sometimes machines can even be trained to detect structures not visible to humans.
Currently the process is cost-intensive and/or a faster decision creates substantial value
A (trained) human could make a good decision mainly based on visual information
There is enough data available (as a rule of thumb: 200 - 2.000 images. This, of course, is highly dependent on the use-case)
In our experience, only by combining knowhow of internal operations with machine vision expertise, projects can be framed well. Feel free to approach us with questions, especially whether we deem your project to be technically feasible.
Together we discuss your process automation projects along three different dimensions: cost savings, strategic value and technical feasibility. After settling for a specific project, we put special emphasis on the needs of the end users.
We are an experienced team of machine learners. Our algorithms find complicated patterns in unstructured, mostly visual and text data. Once detected, these patterns are the basis for the automation of the underlying process.
We make a point of integrating our customers in the project's code repository as well as in weekly progress meetings. Agility, clean code and a modular program structure help us to deliver easy-to-maintain software, that simply works.
Creative solutions enabled us to automate the process of planning solar systems
See how machine learning can be used to stop environmental destruction
AI-supported optical defect detection for semiconductor laser production
We automated the detection of certain cloud structures for Deutscher Wetterdienst (DWD)
An image segmentation algorithm that supports sustainable city planning
We simulate internet traffic and bidding scenarios to predict the reach of advertising campaigns.
By William Clemens (PhD) • September 26th, 2022
This blog post gives a brief introduction to metric learning. It explains common distance measures and loss functions such as triplet loss.
By Konrad Mundinger • August 29th, 2022
In this blog post, I will give an overview of the underlying basic concepts, common use cases and limitations of recommendation systems. Among other topics, I will discuss content-based and collaborative filtering.
By Emilius Richter • August 2nd, 2022
I present the best (Python) tools for remote sensing and processing of satellite data, based on our practical experience with them at dida. For some I provide application examples including code.
By Madina Kasymova • May 31st, 2022
In this article, we examine how an image caption generation pipeline works. In particular, we look at the attention mechanism - a very promising approach to image captioning.
By Felix Brunner • March 21st, 2022
I will walk you step-by-step through creating a simple annotation tool to easily assign class labels to a set of images. The tool can be written and used within a Jupyter Notebook, making use of ipywidgets.
By David Berscheid • October 6th, 2021
In this article, we will see why many ML Projects do not make it into production, introduce the concepts of model- and data-centric ML, and give examples how we at dida improve projects by applying data-centric techniques.
By Tiago Sanona • August 19th, 2021
Following the paper by Garnot et al., we utilize an altered version of the Transformer to classify crop fields from time series of satellite images. With this, we achieve better results than traditional methods (e. g. random forests) and with less resources than recurrent networks.
Machine Learning Project Lead
Machine Learning Consultant