M.Sc. · Institute of Information Processing (TNT), Leibniz University Hannover
Robert Zimmermann studied Business Administration and Engineering at Leibniz University Hannover, specialising in Operations and Information Management, Digitalisation, and Automation. He received his M.Sc. in January 2026 with a master’s thesis at the Institute of Information Processing (TNT) on interpretability methods for vision transformers — work that directly led to the DINO-QPM paper (CVPRW 2026).
Since February 2026 he has been continuing this line of research as a PhD student at the same institute, focusing on the interpretability of visual foundation models.
Our paper DINO-QPM: Adapting Visual Foundation Models for Globally Interpretable Image Classification has been accepted at the 5th Explainable AI for Computer Vision (XAI4CV) Workshop at CVPR 2026.
DINO-QPM: Adapting Visual Foundation Models for Globally Interpretable Image Classification
Robert Zimmermann, Thomas Norrenbrock, and Bodo Rosenhahn
In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops — 5th Explainable AI for Computer Vision (XAI4CV), 2026
Although visual foundation models like DINOv2 provide state-of-the-art performance as feature extractors, their complex, high-dimensional representations create substantial hurdles for interpretability. This work proposes DINO-QPM, which converts these powerful but entangled features into contrastive, class-independent representations that are interpretable by humans. DINO-QPM is a lightweight interpretability adapter that pursues globally interpretable image classification, adapting the Quadratic Programming Enhanced Model (QPM) to operate on strictly frozen DINO backbones. While classification with visual foundation models typically relies on the CLS token, we deliberately diverge from this standard. By leveraging average-pooling, we directly connect the patch embeddings to the model’s features and therefore enable spatial localisation of DINO-QPM’s globally interpretable features within the input space. Furthermore, we apply a sparsity loss to minimise spatial scatter and background noise, ensuring that explanations are grounded in relevant object parts. With DINO-QPM we make the level of interpretability of QPM available as an adapter while exceeding the accuracy of DINOv2 linear probe. Evaluated through an introduced Plausibility metric and other interpretability metrics, extensive experiments demonstrate that DINO-QPM is superior to other applicable methods for frozen visual foundation models in both classification accuracy and explanation quality.
@inproceedings{zimmermann2026dinoqpm,title={DINO-QPM: Adapting Visual Foundation Models for Globally Interpretable Image Classification},author={Zimmermann, Robert and Norrenbrock, Thomas and Rosenhahn, Bodo},booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops — 5th Explainable AI for Computer Vision (XAI4CV)},year={2026},}