Adam Bielski

I obtained my PhD degree in 2024 from the Computer Vision Group at the University of Bern, supervised by Prof. Paolo Favaro.

My research interests lie in unsupervised machine learning in computer vision, including generative models and unsupervised image segmentation.

I studied Telecommunications and Computer Science at the Warsaw University of Technology, where I obtained my bachelor’s degree in 2013 and master’s degree in 2015. My theses focused on speaker recognition using machine learning models.

Before starting my PhD, I gained over 5 years of experience working in the industry on machine learning-related projects, including roles at Samsung R&D, Tooploox, and SnapRapid. During my PhD, I completed an internship at Apple, where I worked on visual language models.

news

May 2, 2024	I obtained my PhD degree from the University of Bern.
May 2, 2023	Starting my new position as a Research Intern at Apple Zurich.
Sep 15, 2022	Our paper “MOVE: Unsupervised Movable Object Segmentation and Detection” accepted to NeurIPS 2022!

Publications

MOVE: Unsupervised Movable Object Segmentation and Detection

Adam Bielski, and Paolo Favaro

Advances in Neural Information Processing Systems (NeurIPS) 2022

Abs arXiv HTML PDF

We introduce MOVE, a novel method to segment objects without any form of supervision. MOVE exploits the fact that foreground objects can be shifted locally relative to their initial position and result in realistic (undistorted) new images. This property allows us to train a segmentation model on a dataset of images without annotation and to achieve state of the art (SotA) performance on several evaluation datasets for unsupervised salient object detection and segmentation. In unsupervised single object discovery, MOVE gives an average CorLoc improvement of 7.2% over the SotA, and in unsupervised class-agnostic object detection it gives a relative AP improvement of 53% on average. Our approach is built on top of self-supervised features (e.g. from DINO or MAE), an inpainting network (based on the Masked AutoEncoder) and adversarial training.
KOALA: A Kalman Optimization Algorithm with Loss Adaptivity

Aram Davtyan, Sepehr Sameni, Llukman Cerkezi, and 3 more authors

Proceedings of the AAAI Conference on Artificial Intelligence 2022

arXiv PDF
Generative Adversarial Learning via Kernel Density Discrimination

Abdelhak Lemkhenter, Adam Bielski, Alp Eren Sari, and 1 more author

arXiv preprint arXiv:2107.06197 2021

arXiv PDF
Emergence of Object Segmentation in Perturbed Generative Models

Adam Bielski, and Paolo Favaro

Advances in Neural Information Processing Systems (NeurIPS), Accepted as Spotlight 2019

Abs arXiv PDF Code

We introduce a novel framework to build a model that can learn how to segment objects from a collection of images without any human annotation. Our method builds on the observation that the location of object segments can be perturbed locally relative to a given background without affecting the realism of a scene. Our approach is to first train a generative model of a layered scene. The layered representation consists of a background image, a foreground image and the mask of the foreground. A composite image is then obtained by overlaying the masked foreground image onto the background. The generative model is trained in an adversarial fashion against a discriminator, which forces the generative model to produce realistic composite images. To force the generator to learn a representation where the foreground layer corresponds to an object, we perturb the output of the generative model by introducing a random shift of both the foreground image and mask relative to the background. Because the generator is unaware of the shift before computing its output, it must produce layered representations that are realistic for any such random perturbation. Finally, we learn to segment an image by defining an autoencoder consisting of an encoder, which we train, and the pre-trained generator as the decoder, which we freeze. The encoder maps an image to a feature vector, which is fed as input to the generator to give a composite image matching the original input image. Because the generator outputs an explicit layered representation of the scene, the encoder learns to detect and segment objects. We demonstrate this framework on real images of several object categories.
Understanding multimodal popularity prediction of social media videos with self-attention

Adam Bielski, and Tomasz Trzcinski

IEEE Access 2018

HTML PDF
Pay Attention to Virality: understanding popularity of social media videos with the attention mechanism

Adam Bielski, and Tomasz Trzcinski

Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), Accepted as oral presentation 2018

arXiv PDF
I Know How You Feel: Emotion Recognition with Facial Landmarks

Ivona Tautkute, Tomasz Trzcinski, and Adam Bielski

2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW) 2018

arXiv PDF
Recognizing Emotions with EmotionalDAN

Ivona Tautkute, Tomasz Trzciński, and Adam Bielski

International Symposium Computational Modeling of Objects Represented in Images 2018

HTML
Extracting textual overlays from social media videos using neural networks

Adam Słucki, Tomasz Trzciński, Adam Bielski, and 1 more author

International Conference on Computer Vision and Graphics 2018

HTML
SocialML: machine learning for social media video creators

Tomasz Trzcinski, Adam Bielski, Paweł Cyrta, and 1 more author

NIPS 2017 Workshop on Machine Learning for Creativity and Design 2017

arXiv PDF