Matthias Bethge

EbC learns equivariant embeddings from observation pairs without relying on group-specific inductive biases, with theoretical …

RDumb is a simple baseline that periodically resets the model to its pretrained state, yet outperforms state-of-the-art continual …

Unsupervised object learning from videos using the Common Fate Principle, decomposing the problem into motion segmentation, generative …

Contrastive learning with the InfoNCE objective can recover the ground truth latent factors underlying the data.

Self-learning techniques like entropy minimization and pseudo-labeling are simple and effective at improving model performance under …

ImageNet pretraining significantly improves out-of-domain robustness for pose estimation across architectures and species.

Adapting batch norm statistics of trained computer vision models considerably increases robustness at test-time.

We show that good target performance can be achieved on domain adaptation tasks by adapting only the normalization statistics and …

An open source toolbox providing a unified implementation of state-of-the-art methods for transfer learning, semi-supervised learning …