Dina Bashkirova

I am a PhD candidate at the Computer Science Department of Boston University advised by Professor Kate Saenko. My research is focused on semantically-consistent unsupervised pixel-level domain adaptation. I am also interested in label-efficient object localization and semantic segmentation, as well as applications of AI and Computer Vision for Good.

Prior to that, I received my BSc and MSc in Computer Science at Kazan Federal University.

dbash [at] bu [dot] edu  /  CV  /  Google Scholar

profile photo
Research

Currently, I am working on the cross-domain disentanglement in image-to-image translation and weakly-supervised object localization under domain shift. In addition to that, I am working on the label-efficient object segmentation in a challenging industrial waste sorting setup.

clean-usnob Disentangled Unsupervised Image Translation via Restricted Information Flow
Ben Usman*, Dina Bashkirova*, Kate Saenko
(in submission) 2021
arxiv / bib

We propose a new many-to-many image translation method that infers which attributes are domain-specific from data by constraining information flow through the network using translation honesty losses and a penalty on the capacity of the domain-specific embedding, and does not rely on hard-coded inductive architectural biases.

clean-usnob ZeroWaste Dataset: Towards Deformable Object Segmentation in Cluttered Scenes
Dina Bashkirova, Mohamed Abdelfattah, Ziliang Zhu, James Akl, Fadi Alladkani, Ping Hu, Vitaly Ablavsky, Berk Calli, Sarah Adel Bargal Kate Saenko
CVPR 2022
arxiv / project page / bib / code

We present the first in-the-wild object segmentation dataset for industrial waste sorting for fully-, semi- and weakly-supervised setups. Our ZeroWaste datasets presents a challenging computer vision task of semantic segmentation of extremely cluttered scenes with highly deformable and translucent objects.

clean-usnob Evaluation of Correctness in Unsupervised Many-to-Many Image Translation
Dina Bashkirova, Ben Usman, Kate Saenko
WACV 2022
arxiv / github / bib

We propose an evaluation protocol for the disentanglement quality of unsupervised many-to-many image translation (UMI2I) methods. We show that modern UMI2I methods fail to correctly disentangle the domain-specific from shared factors and mostly rely on their corresponding inductive biases to determine which factors should be changed after translation.

clean-usnob MixtureEnsembles: Leveraging Parameter Sharing for Efficient Ensembles
Piotr Teterwak, Nikoli Dryden, Dina Bashkirova, Kate Saenko Bryan A. Plummer
Under Review, Preprint coming soon. 2021

We propose a parameter-efficient ensembling method with shape-shifter networks.

clean-usnob Compositional Models: Multi-Task Learning and Knowledge Transfer with Modular Networks
Andrey Zhmoginov, Mark Sandler, Dina Bashkirova.
arXiv 2020
arxiv / github / bib

We propose a novel modular model architecture that allows parameter sharing and reuse for various computer vision tasks, such as multi-task learning, knowledge transfer and domain adaptation.

clean-usnob Adversarial Self-Defense for Cycle-Consistent GANs
Dina Bashkirova, Ben Usman, Kate Saenko
NeurIPS 2019
arxiv / github / project page / poster / proceedings / bib

We show that cycle-consistent models perform a self-adversarial attack by embedding low-amplitude structured noise into intermediate generated images to reconstruct input images perfectly. We propose two techniques that prevent this kind of "cheating" and show that defending against such self-adversarial attacks improves the translation quality.

clean-usnob Unsupervised video-to-video translation
Dina Bashkirova, Ben Usman, Kate Saenko
arXiv 2018
arxiv / github / volumetric data / bib

We propose a spatiotemporal extention of CycleGAN and show when it performs better then per-frame translation on two novel unsupervised video-to-video translation benchmarks including a novel CT-to-MRI volumetric medical domain.

clean-usnob Fast L1 Gauss transforms for edge-aware image filtering
Dina Bashkirova, Shin Yoshizawa, Roustam Latypov, Hideo Yokota
Proceedings of ISP RAS 2017
arxiv / bib

We propose a novel approximation method for fast Gaussian convolution of two-dimensional uniform point sets that involves L1 distance metric for Gaussian function and domain splitting approach to achieve fast computation.

Service

Grader for CS480 Introduction to Computer Graphics in 2018, CS542 Machine Learning in 2020 and 2021.

Reviewer for NeurIPS 2018-2021, ICCV 2021, ICLR 2020, 2021, CVPR 2020, WACV 2020, 2021, NeurIPS DistShift Workshop 2021. Helped running VisDA challenge at NeurIPS21.


Template adapted from Jon Barron's homepage.