Searcharxiv⌕ Search

arXiv subjects

Hojin Kim

Publications and source records attributed to Hojin Kim.

22 records · Page 2Linked to original sources

Offline-to-Online Knowledge Distillation for Video Instance Segmentation

In this paper, we present offline-to-online knowledge distillation (OOKD) for video instance segmentation (VIS), which transfers a wealth of video knowledge from an offline model to an online model for consistent prediction. Unlike previous methods that having adopting either an online or offline model, our single online model takes advantage of both models by distilling offline knowledge. To transfer knowledge correctly, we propose query filtering and association (QFA), which filters irrelevant queries to exact instances. Our KD with QFA increases the robustness of feature matching by encoding object-centric features from a single frame supplemented by long-range global information. We also propose a simple data augmentation scheme for knowledge distillation in the VIS task that fairly transfers the knowledge of all classes into the online model. Extensive experiments show that our method significantly improves the performance in video instance segmentation, especially for challenging datasets including long, dynamic sequences. Our method also achieves state-of-the-art performance on YTVIS-21, YTVIS-22, and OVIS datasets, with mAP scores of 46.1%, 43.6%, and 31.1%, respectively.

cs.CV↗

Hopf algebras and multiple zeta values in positive characteristic

Multiples zeta values (MZV's for short) in positive characteristic were introduced by Thakur as analogues of classical multiple zeta values of Euler. In this paper we give a systematic study of algebraic structures of MZV's in positive characteristic. We construct both the stuffle algebra and the shuffle algebra of these MZV's and equip them with algebra and Hopf algebra structures. In particular, we completely solve a problem suggested by Deligne and Thakur \cite{Del17} in 2017 and establish Shi's conjectures \cite{Shi18}. The construction of the stuffle algebra is based on our recent work \cite{IKLNDP22}.

math.NT↗

On the common zeros of quasi-modular forms for $Γ_0^+(N)$ of level $N=1,2,3$

In this paper, we study common zeros of the iterated derivatives of the Eisenstein series for $Γ_0^+(N)$ of level $N=1,2$ and $3$, which are quasi-modular forms. More precisely, we investigate the common zeros of quasi-modular forms, and prove that all the zeros of the iterated derivatives of the Eisenstein series $\frac{d^m E_k^{(N)}(τ)}{dτ^m}$ of weight $k=2,4,6$ for $Γ_0^+(N)$ of level $N=2,3$ are simple by generalizaing the results of Meher \cite{MEH} and Gun and Oesterlé \cite{SJ20} for SL$_2(\mathbb{Z})$.

math.NT↗

Designing Stress-Adaptive Dense Suspensions using Dynamic Covalent Chemistry

The non-Newtonian behaviors of dense suspensions are central to their use in technological and industrial applications and arise from a network of particle-particle contacts that dynamically adapts to imposed shear. Reported herein are studies aimed at explor-ing how dynamic covalent chemistry between particles and the polymeric solvent can be used to tailor such stress-adaptive contact networks leading to their unusual rheological behaviors. Specifically, a room temperature dynamic thia-Michael bond is employed to rationally tune the equilibrium constant (Keq) of the polymeric solvent to the particle interface. It is demonstrated that low Keq leads to shear thinning while high Keq produces antithixotropy, a rare phenomenon where the viscosity increases with shearing time. It is proposed that an increase in Keq increases the polymer graft density at the particle surface and that antithixotropy primar-ily arises from partial debonding of the polymeric graft/solvent from the particle surface and the formation of polymer bridges between particles. Thus, the implementation of dynamic covalent chemistry provides a new molecular handle with which to tailor the macroscopic rheology of suspensions by introducing programable time dependence. These studies open the door to energy absorbing materials that not only sense mechanical inputs and adjust their dissipation as a function of time or shear rate but can switch between these two modalities on demand.

cond-mat.soft↗