Leonard Salewski

Leonard Salewski is a Ph.D. candidate in the International Max-Planck Research School for Intelligent Systems (IMPRS-IS) under the joint supervision of Prof. Zeynep Akata and Prof. Hendrik Lensch


Leonard Salewski
Phone +49 (0)7071 29-60070794

Department of Computer Science

Maria-von-Linden-Str. 6

72076 Tübingen, Germany

Room 40-7/A18


Zero-shot Translation of Attention Patterns in VQA Models to Natural Language.
Leonard Salewski, A. Sophia Koepke, Hendrik P. A. Lensch and Zeynep Akata
To appear in: German Conference on Pattern Recognition, 2023


In-Context Impersonation Reveals Large Language Models' Strengths and Biases.
Leonard Salewski, Stephan Alaniz, Isabel Rio-Torto, Eric Schulz and Zeynep Akata
ArXiv abs/2305.14930, 2023


Diverse Video Captioning by Adaptive Spatio-temporal Attention.
Zohreh Ghaderi, Leonard Salewski and Hendrik P. A. Lensch
German Conference on Pattern Recognition, 2022


CLEVR-X: A visual reasoning dataset for natural language explanations.
Leonard Salewski, A. Sophia Koepke, Hendrik P. A. Lensch and Zeynep Akata
Springer Lecture Notes on Artificial Intelligence, 2022
Paper | Project page | Code
This was also presented at the CVPR 2022 Workshop on Explainable AI for Computer Vision (XAI4CV).


e-ViL: A Dataset and Benchmark for Natural Language Explanations in Vision-Language Tasks.
Maxime Kayser, Oana-Maria Camburu, Leonard Salewski, Cornelius Emde, Virginie Do, Zeynep Akata and Thomas Lukasiewicz
IEEE International Conference of Computer Vision, ICCV 2021
Paper | Code


Relational Generalized Few-Shot Learning.
Xiahan Shi, Leonard Salewski, Martin Schiegg, Zeynep Akata and Max Welling
British Machine Vision Conference, 2020
This publication is the result of my master thesis.


For up-to-date information please also check: Semantic Scholar or Google Scholar.

Research interests

  • Natural Language Explanations for Computer Vision Problems
  • Properties of Large Language Models
  • Deep learning for vision-language tasks

Additionally I am working on scholarGPT, an academic chatbot, that does not hallucinate its sources, but instead gives reliable and traceable answers based on >2.25M arXiv pre-prints. Further applications for scholarGPT, can be found in law, journalism and education.


  • TPAMI 2021 / 2023
  • MULA 2022
  • CVPR 2023 (Emergency)
  • IJCV 2023 (2x)
  • BMVC 2023 (Emergency)


Master Thesis