arXiv subjects

Minchul Kim

Publications and source records attributed to Minchul Kim.

26 records · Page 2Linked to original sources

arXiv2023

AdaFace: Quality Adaptive Margin for Face Recognition

Recognition in low quality face datasets is challenging because facial attributes are obscured and degraded. Advances in margin-based loss functions have resulted in enhanced discriminability of faces in the embedding space. Further, previous studies have studied the effect of adaptive losses to assign more importance to misclassified (hard) examples. In this work, we introduce another aspect of adaptiveness in the loss function, namely the image quality. We argue that the strategy to emphasize misclassified samples should be adjusted according to their image quality. Specifically, the relative importance of easy or hard samples should be based on the sample's image quality. We propose a new loss function that emphasizes samples of different difficulties based on their image quality. Our method achieves this in the form of an adaptive margin function by approximating the image quality with feature norms. Extensive experiments show that our method, AdaFace, improves the face recognition performance over the state-of-the-art (SoTA) on four datasets (IJB-B, IJB-C, IJB-S and TinyFace). Code and models are released in https://github.com/mk-minchul/AdaFace.

cs.CV↗

arXiv2023

Cluster and Aggregate: Face Recognition with Large Probe Set

Feature fusion plays a crucial role in unconstrained face recognition where inputs (probes) comprise of a set of $N$ low quality images whose individual qualities vary. Advances in attention and recurrent modules have led to feature fusion that can model the relationship among the images in the input set. However, attention mechanisms cannot scale to large $N$ due to their quadratic complexity and recurrent modules suffer from input order sensitivity. We propose a two-stage feature fusion paradigm, Cluster and Aggregate, that can both scale to large $N$ and maintain the ability to perform sequential inference with order invariance. Specifically, Cluster stage is a linear assignment of $N$ inputs to $M$ global cluster centers, and Aggregation stage is a fusion over $M$ clustered features. The clustered features play an integral role when the inputs are sequential as they can serve as a summarization of past features. By leveraging the order-invariance of incremental averaging operation, we design an update rule that achieves batch-order invariance, which guarantees that the contributions of early image in the sequence do not diminish as time steps increase. Experiments on IJB-B and IJB-S benchmark datasets show the superiority of the proposed two-stage paradigm in unconstrained face recognition. Code and pretrained models are available in https://github.com/mk-minchul/caface

cs.CV↗

arXiv2022

Controllable and Guided Face Synthesis for Unconstrained Face Recognition

Although significant advances have been made in face recognition (FR), FR in unconstrained environments remains challenging due to the domain gap between the semi-constrained training datasets and unconstrained testing scenarios. To address this problem, we propose a controllable face synthesis model (CFSM) that can mimic the distribution of target datasets in a style latent space. CFSM learns a linear subspace with orthogonal bases in the style latent space with precise control over the diversity and degree of synthesis. Furthermore, the pre-trained synthesis model can be guided by the FR model, making the resulting images more beneficial for FR model training. Besides, target dataset distributions are characterized by the learned orthogonal bases, which can be utilized to measure the distributional similarity among face datasets. Our approach yields significant performance gains on unconstrained benchmarks, such as IJB-B, IJB-C, TinyFace and IJB-S (+5.76% Rank1).

cs.CV↗

arXiv2020

Information-Theoretic Privacy in Federated Submodel learning

We consider information-theoretic privacy in federated submodel learning, where a global server has multiple submodels. Compared to the privacy considered in the conventional federated submodel learning where secure aggregation is adopted for ensuring privacy, information-theoretic privacy provides the stronger protection on submodel selection by the local machine. We propose an achievable scheme that partially adopts the conventional private information retrieval (PIR) scheme that achieves the minimum amount of download. With respect to computation and communication overhead, we compare the achievable scheme with a naive approach for federated submodel learning with information-theoretic privacy.

cs.IT↗

arXiv2020

Learning Visual Context by Comparison

Finding diseases from an X-ray image is an important yet highly challenging task. Current methods for solving this task exploit various characteristics of the chest X-ray image, but one of the most important characteristics is still missing: the necessity of comparison between related regions in an image. In this paper, we present Attend-and-Compare Module (ACM) for capturing the difference between an object of interest and its corresponding context. We show that explicit difference modeling can be very helpful in tasks that require direct comparison between locations from afar. This module can be plugged into existing deep learning models. For evaluation, we apply our module to three chest X-ray recognition tasks and COCO object detection & segmentation tasks and observe consistent improvements across tasks. The code is available at https://github.com/mk-minchul/attend-and-compare.

cs.CV↗

arXiv2019

Private Secure Coded Computation

We introduce a variation of coded computation that ensures data security and master's privacy against workers, which is referred to as private secure coded computation. In private secure coded computation, the master needs to compute a function of its own dataset and one of the datasets in a library exclusively shared by external workers. After recovering the desired function, the workers must not be able to know which dataset in the library was desired by the master or obtain any information about the master's own data. We propose a private secure coded computation scheme for matrix multiplication, namely private secure polynomial codes, based on private polynomial codes for private coded computation. In simulations, we show that the private secure polynomial codes achieves better computation time than private polynomial codes modified for private secure coded computation.

cs.IT↗

arXiv2018

Private Coded Computation for Machine Learning

In a distributed computing system for the master-worker framework, an erasure code can mitigate the effects of slow workers, also called stragglers. The distributed computing system combined with coding is referred to as coded computation. We introduce a variation of coded computation that protects the master's privacy from the workers, which is referred to as private coded computation. In private coded computation, the master needs to compute a function of its own dataset and one of the datasets in a library exclusively shared by the external workers. After the master recovers the result of the desired function through coded computation, the workers should not know which dataset in the library was desired by the master, which implies that the master's privacy is protected. We propose a private coded computation scheme for matrix multiplication, namely private polynomial codes, based on polynomial codes for conventional coded computation. As special cases of private polynomial codes, we propose private one-shot polynomial codes and private asynchronous polynomial codes. Whereas the private one-shot polynomial code achieves a lower communication load from the master to each worker, the private asynchronous polynomial code achieves faster computation than private one-shot polynomial codes. In terms of computation time and communication load, we compare private one-shot polynomial codes and private asynchronous polynomial codes with a conventional robust private information retrieval scheme which can be directly applied to coded computation.

cs.IT↗

arXiv2018

Repair of Multiple Descriptions on Distributed Storage

In multiple descriptions on distributed storage, a source is stored in a shared fashion on multiple servers. When a subset of servers are contacted, the source should be estimated with a certain maximum distortion depending on the number of servers. The problem considered in this paper is how to restore the system operation when one of the servers fail and a new server replaces it, that is, repair. The requirement is that the distortions in the restored system should be no more than in the original system. The question is how many extra bits are needed for repair. We find the optimum solution for a two server problem in the Gaussian case, and an achievable rate for general $n$ nodes. One conclusion is that it is necessary to design the multiple description codes with repair in mind; just using an existing multiple description code results in unnecessary high repair rates.

cs.IT↗