Manoj Kumar

selected publications

Frozen Feature Augmentation for Few-Shot Image Classification

Andreas Bär, Neil Houlsby, Mostafa Dehghani, and 1 more author

CVPR 2024,

Abs arXiv Website

A large study of image augmentations but in frozen pretrained embedding space Simple augmentations like brightness, contrast and posterize work best.
Image captioners are scalable vision learners too

Michael Tschannen, Manoj Kumar, Andreas Steiner, and 3 more authors

NeurIPS Oral 2024,

Abs arXiv Code Website

Image encoders trained with captioning pretraining are competitive with contrastive pretraining.
Dual PatchNorm

Manoj Kumar, Mostafa Dehghani, and Neil Houlsby

TMLR 2023,

Abs arXiv Website

Adding a LayerNorm before and after the ViT embedding layer improves performance on a number of benchmarks.
Scaling Vision Transformers to 22 Billion Parameters

Mostafa Dehghani, Josip Djolonga, Basil Mustafa, and 8 more authors

ICML Oral 2023,

Abs arXiv Website

title.
Do better ImageNet classifiers assess perceptual similarity better?

Manoj Kumar, Neil Houlsby, Nal Kalchbrenner, and 1 more author

TMLR 2022,

Abs arXiv Blog Code Website

Scaling down ImageNet classifiers leads to state-of-the-art results on perceptual similarity!
Deep learning for twelve hour precipitation forecasts

Lasse Espeholt, Shreya Agrawal, Casper Sønderby, and 8 more authors

Nature Communications 2022,

Abs arXiv Blog Website

Neural networks equipped with large spatial contexts can make highly effective precipitation rate predictions
Colorization Transformer

Manoj Kumar, Dirk Weissenborn, and Nal Kalchbrenner

ICLR 2021,

Abs arXiv Code Website

3 stage transformer for automaged 256x256 image colorization.
VideoFlow: A Conditional Flow-Based Model for Stochastic Video Generation

Manoj Kumar, Mohammad Babaeizadeh, Dumitru Erhan, and 4 more authors

ICLR 2020,

Abs arXiv Code Website

First application of normalizing flows for video prediction.