How do you optimize the AI Image Enhancer, so the processed images will be improved? That’s the project that our data scientist Anders Launer Bæk is working on and his work is a crucial part of improving the AI Image Enhancer so the processed images should be closer to the editor’s final output.

How can we improve our AI Image Enhancer with semantic segmentation? That is the question currently being investigated by our data scientist Anders. The overall objective to this question is to explore methods producing a rich information for a given image. 

By combining these new methods with our AI Image Enhancer, we hope to learn a variety of transformations which will result in images closer to the editor’s final output.  

Based upon the findings from our Ph.D project AIERE, by Juan Francisco Marin Vega. The project is driven by our curiosity to answer the above-mentioned question. It also seeks to study the recent literature within semantic segmentation for future projects in our portfolio. 

The project applies both open source and “in house” software. We are very fond of the FastAI framework as well as the PyTorch deep learning engine for model experimentation. Currently we are using our digital asset management system to ensure reproducibility throughout our experiments. 

We are tracking the explorative phase of this project with WANDB.ai. So far, our results only validate the fact that the predictive power of deep learning models is bounded by the quality of the data foundation. This challenge is accommodated in-house by our excellent colleagues in Hanoi, continuously delivering high-quality data. 

As we move forward, we are focusing on fusing semantic segmentation into the AI Image Enhancer, a challenge we foresee overcoming in the near future.