The Image Sorting Experiment

Maximizing Image Viewing Efficiency: How Visual Sorting Can Help

TLDR: In January 2022, we – the Visual Computing Group at HTW Berlin – conducted an experiment to evaluate image sorting. It was shown that images in sorted arrangements are found much more quickly. Our new measure for evaluating image sorting proved to be significantly better than those usually used to describe the perceived sorting quality by humans. Additionally, our proposed sorting methods were able to generate high-quality image sorting much more efficiently compared to other methods.

More than 2000 participants took part in our experiment, and we would like to thank them again here. The published article ( on the results of the experiment  may be difficult to understand for non-specialists. Therefore, we will attempt to summarize the motivation, implementation, and results of the experiment in an understandable way here.


Das Bildsortierungs-Experiment

TLDR: Im Januar 2022 führten wir – die Visual Computing Gruppe der HTW Berlin – ein Experiment zur Bewertung von Bildsortierungen durch. Dabei konnte gezeigt werden, dass Bilder in sortierten Anordnungen viel schneller gefunden werden. Unser neues Maß zur Bewertung von Bildsortierungen erwies sich als deutlich besser als die üblicherweise verwendeten, um die von Menschen wahrgenommene Sortierqualität zu beschreiben. Zudem konnten unsere vorgeschlagenen Sortierverfahren im Vergleich zu anderen Verfahren viel effizienter qualitativ hochwertige Bildsortierungen erzeugen.

Mehr als 2000 Teilnehmer haben an unserem Experiment teilgenommen, denen wir hier noch einmal danken möchten. Die Ergebnisse wurden in einem Fachartikel ( ) veröffentlicht, der jedoch für Nicht-Spezialisten möglicherweise schwer zu verstehen ist. Daher versuchen wir hier, die Motivation, die Durchführung und die Ergebnisse des Experiments verständlich zusammenzufassen.


Converting Tensorflow graphs into TensorRT networks


A few weeks ago we needed to convert one of our own Tensorflow graphs into a TensorRT network. As many of you probably know, there are a few options to accomplish this, like the Tensorflow to UFF and UFF to TensorRT parser or the Tensorflow to ONNX and ONNX to TensorRT parser. When trying the first approach the following error message was one of many we encountered: UffParser: Validator error: slice_9-26_9-26: Unsupported operation Slice. Some of the problems are circumventable but in the end we had to abandon the UFF to TensorRT parser, since it is full of bugs and closed source. The ONNX way seemed more promising since its intermediate format was visualisable and changeable. Unfortunately the packages provided by Anaconda and PyPI were flawed and fixing the C++ source code felt like a lot of work. Especially since the python API of TensorRT to construct networks looked clean and had all operations we needed.