Object Re-identification Internship

One of the projects we are working on centers around checking medication using A.I., as a means to fulfil the so-called 4-eyes principle.
One of the challenges in this project is to check whether a set of pills visible in an image are the same as pills as were visible in an image taken on the previous day. This means a solution will need to be investigated to re-identify objects between multiple images, even if aspects like background, illumination, camera angle can be completely different from one image to another. The goal of such an algorithm is to notify users of unexpected changes in the amount and composition of their medication.

The challenge of this internship is to investigate multiple approaches for object re-identification. Possible solutions could lie in the direction of mapping object features onto a global feature space and applying classic search algorithms to identify matches. On the other hand a deep-learning based solution might also be very much worth investigating.

You will be given large freedom in finding the best approach(es). For inspiration you could already have a look at:

• Surveys on object re-identification:

  • https://www.intechopen.com/chapters/67472

  • https://arxiv.org/pdf/2001.04193.pdf

• Clustering & dimensionality reduction:

  • t-SNE

  • DBSCAN

  • PCA

• Deep Learning based methods:

  • Metric Learning

  • Triplet Loss

  • Siamese networks + contrastive loss

Note that we already have a custom, high-quality data set with segmentation annotations to work with.

Can you create an algorithm that can tell us that the set of pills in the left image is not the same as the set of pills in the middle image, but at the same time tell us that the pills in the middle image are the same as the pills in the rightmost image?

About Clockworks

Clockworks is a fast growing, Rotterdam based company specialised in the development of smart data capture solutions on the cutting edge of Artificial Intelligence and Computer Vision. Our motto being ‘for the love of pixels’, we have a strong passion for technology and developing innovative and creative software solutions using the power of pixels and AI. Our growing team consists of 10 people, most of which have technology backgrounds in fields such as artificial intelligence, computer science and applied physics.

What are we looking for?

• For this internship you should have some background knowledge into data science and/or artificial intelligence

• Affinity with computer vision and pixels in general is recommended but not required

• You are studying at a Dutch university (WO/HBO) in a field such as computer science, artificial intelligence, data science, etc.

Practical information

• The duration of the project is flexible, but 3-6 months (full-time) is preferred

• Starting date: flexible, as soon as possible.

• Our office is based in Rotterdam

  Goudsesingel, walking distance from subway Oostplein and train station Rotterdam Blaak

• Internship allowance applicable

• Having your own laptop is recommended