arXiv subjects

Dongwon Kim

Publications and source records attributed to Dongwon Kim.

At least 37 records · Page 2Linked to original sources

arXiv2025

Robust Bayesian Optimization via Localized Online Conformal Prediction

Bayesian optimization (BO) is a sequential approach for optimizing black-box objective functions using zeroth-order noisy observations. In BO, Gaussian processes (GPs) are employed as probabilistic surrogate models to estimate the objective function based on past observations, guiding the selection of future queries to maximize utility. However, the performance of BO heavily relies on the quality of these probabilistic estimates, which can deteriorate significantly under model misspecification. To address this issue, we introduce localized online conformal prediction-based Bayesian optimization (LOCBO), a BO algorithm that calibrates the GP model through localized online conformal prediction (CP). LOCBO corrects the GP likelihood based on predictive sets produced by LOCBO, and the corrected GP likelihood is then denoised to obtain a calibrated posterior distribution on the objective function. The likelihood calibration step leverages an input-dependent calibration threshold to tailor coverage guarantees to different regions of the input space. Under minimal noise assumptions, we provide theoretical performance guarantees for LOCBO's iterates that hold for the unobserved objective function. These theoretical findings are validated through experiments on synthetic and real-world optimization tasks, demonstrating that LOCBO consistently outperforms state-of-the-art BO algorithms in the presence of model misspecification.

cs.LG↗

arXiv2024

Improving Text-based Person Search via Part-level Cross-modal Correspondence

Text-based person search is the task of finding person images that are the most relevant to the natural language text description given as query. The main challenge of this task is a large gap between the target images and text queries, which makes it difficult to establish correspondence and distinguish subtle differences across people. To address this challenge, we introduce an efficient encoder-decoder model that extracts coarse-to-fine embedding vectors which are semantically aligned across the two modalities without supervision for the alignment. There is another challenge of learning to capture fine-grained information with only person IDs as supervision, where similar body parts of different individuals are considered different due to the lack of part-level supervision. To tackle this, we propose a novel ranking loss, dubbed commonality-based margin ranking loss, which quantifies the degree of commonality of each body part and reflects it during the learning of fine-grained body part details. As a consequence, it enables our method to achieve the best records on three public benchmarks.

cs.CV↗

arXiv2024

1.58-bit FLUX

We present 1.58-bit FLUX, the first successful approach to quantizing the state-of-the-art text-to-image generation model, FLUX.1-dev, using 1.58-bit weights (i.e., values in {-1, 0, +1}) while maintaining comparable performance for generating 1024 x 1024 images. Notably, our quantization method operates without access to image data, relying solely on self-supervision from the FLUX.1-dev model. Additionally, we develop a custom kernel optimized for 1.58-bit operations, achieving a 7.7x reduction in model storage, a 5.1x reduction in inference memory, and improved inference latency. Extensive evaluations on the GenEval and T2I Compbench benchmarks demonstrate the effectiveness of 1.58-bit FLUX in maintaining generation quality while significantly enhancing computational efficiency.

cs.CV↗

arXiv2024

Bootstrapping Top-down Information for Self-modulating Slot Attention

Object-centric learning (OCL) aims to learn representations of individual objects within visual scenes without manual supervision, facilitating efficient and effective visual reasoning. Traditional OCL methods primarily employ bottom-up approaches that aggregate homogeneous visual features to represent objects. However, in complex visual environments, these methods often fall short due to the heterogeneous nature of visual features within an object. To address this, we propose a novel OCL framework incorporating a top-down pathway. This pathway first bootstraps the semantics of individual objects and then modulates the model to prioritize features relevant to these semantics. By dynamically modulating the model based on its own output, our top-down pathway enhances the representational quality of objects. Our framework achieves state-of-the-art performance across multiple synthetic and real-world object-discovery benchmarks.

cs.CV↗

arXiv2024

PLOT: Text-based Person Search with Part Slot Attention for Corresponding Part Discovery

Text-based person search, employing free-form text queries to identify individuals within a vast image collection, presents a unique challenge in aligning visual and textual representations, particularly at the human part level. Existing methods often struggle with part feature extraction and alignment due to the lack of direct part-level supervision and reliance on heuristic features. We propose a novel framework that leverages a part discovery module based on slot attention to autonomously identify and align distinctive parts across modalities, enhancing interpretability and retrieval accuracy without explicit part-level correspondence supervision. Additionally, text-based dynamic part attention adjusts the importance of each part, further improving retrieval outcomes. Our method is evaluated on three public benchmarks, significantly outperforming existing methods.

cs.CV↗

arXiv2024

Extending CLIP's Image-Text Alignment to Referring Image Segmentation

Referring Image Segmentation (RIS) is a cross-modal task that aims to segment an instance described by a natural language expression. Recent methods leverage large-scale pretrained unimodal models as backbones along with fusion techniques for joint reasoning across modalities. However, the inherent cross-modal nature of RIS raises questions about the effectiveness of unimodal backbones. We propose RISCLIP, a novel framework that effectively leverages the cross-modal nature of CLIP for RIS. Observing CLIP's inherent alignment between image and text features, we capitalize on this starting point and introduce simple but strong modules that enhance unimodal feature extraction and leverage rich alignment knowledge in CLIP's image-text shared-embedding space. RISCLIP exhibits outstanding results on all three major RIS benchmarks and also outperforms previous CLIP-based methods, demonstrating the efficacy of our strategy in extending CLIP's image-text alignment to RIS.

cs.CV↗

arXiv2023

Effect of Resonant Acoustic Powder Mixing on Delay Time of W-KClO4-BaCrO4 Mixtures

This study investigates the impact of resonant acoustic powder mixing on the delay time of the W-KClO4-BaCrO4 (WKB) mixture and its potential implications for powder and material synthesis. Through thermal analysis, an inverse linear relationship was found between thermal conductivity and delay time, allowing us to use thermal conductivity as a reliable proxy for the delay time. By comparing the thermal conductivity of WKB mixtures mixed manually and using acoustic powder mixer, we found that acoustic powder mixing resulted in minimal deviations in thermal conductivity, proving more uniform mixing. Furthermore, DSC analysis and Sestak-Berggren modeling demonstrated consistent reaction dynamics with a constant activation energy as the reaction progressed in samples mixed using acoustic waves. These findings underscore the critical role of uniform powder mixing in enhancing the thermodynamic quality of the WKB mixture and emphasize the importance of developing novel methods for powder and material synthesis.

physics.app-ph↗

arXiv2023

Shatter and Gather: Learning Referring Image Segmentation with Text Supervision

Referring image segmentation, the task of segmenting any arbitrary entities described in free-form texts, opens up a variety of vision applications. However, manual labeling of training data for this task is prohibitively costly, leading to lack of labeled data for training. We address this issue by a weakly supervised learning approach using text descriptions of training images as the only source of supervision. To this end, we first present a new model that discovers semantic entities in input image and then combines such entities relevant to text query to predict the mask of the referent. We also present a new loss function that allows the model to be trained without any further supervision. Our method was evaluated on four public benchmarks for referring image segmentation, where it clearly outperformed the existing method for the same task and recent open-vocabulary segmentation models on all the benchmarks.

cs.CV↗

arXiv2023

Improving Cross-Modal Retrieval with Set of Diverse Embeddings

Cross-modal retrieval across image and text modalities is a challenging task due to its inherent ambiguity: An image often exhibits various situations, and a caption can be coupled with diverse images. Set-based embedding has been studied as a solution to this problem. It seeks to encode a sample into a set of different embedding vectors that capture different semantics of the sample. In this paper, we present a novel set-based embedding method, which is distinct from previous work in two aspects. First, we present a new similarity function called smooth-Chamfer similarity, which is designed to alleviate the side effects of existing similarity functions for set-based embedding. Second, we propose a novel set prediction module to produce a set of embedding vectors that effectively captures diverse semantics of input by the slot attention mechanism. Our method is evaluated on the COCO and Flickr30K datasets across different visual backbones, where it outperforms existing methods including ones that demand substantially larger computation at inference.

cs.CV↗

arXiv2023

Detailed chemical abundances of stars in the outskirts of the Tucana II ultra-faint dwarf galaxy

We present chemical abundances and velocities of five stars between 0.3 kpc to 1.1 kpc from the center of the Tucana II ultra-faint dwarf galaxy (UFD) from high-resolution Magellan/MIKE spectroscopy. We find that every star is deficient in metals (-3.6 < [Fe/H] < -1.9) and in neutron-capture elements as is characteristic of UFD stars, unambiguously confirming their association with Tucana II. Other chemical abundances (e.g., C, iron-peak) largely follow UFD trends and suggest that faint core-collapse supernovae (SNe) dominated the early evolution of Tucana II. We see a downturn in [$α$/Fe] at [Fe/H] $\approx -2.8$, indicating the onset of Type Ia SN enrichment and somewhat extended chemical evolution. The most metal-rich star has strikingly low [Sc/Fe] = $-1.29 \pm 0.48$ and [Mn/Fe] = $-1.33 \pm 0.33$, implying significant enrichment by a sub-Chandrasekhar mass Type Ia SN. We do not detect a radial velocity gradient in Tucana II ($\text{d}v_{\text{helio}}/\text{d}θ_1=-2.6^{+3.0}_{-2.9}$ km s$^{-1}$ kpc$^{-1}$) reflecting a lack of evidence for tidal disruption, and derive a dynamical mass of $M_{1/2} (r_h) = 1.6^{+1.1}_{-0.7}\times 10^6$ M$_{\odot}$. We revisit formation scenarios of the extended component of Tucana II in light of its stellar chemical abundances. We find no evidence that Tucana II had abnormally energetic SNe, suggesting that if SNe drove in-situ stellar halo formation then other UFDs should show similar such features. Although not a unique explanation, the decline in [$α$/Fe] is consistent with an early galactic merger triggering later star formation. Future observations may disentangle such formation channels of UFD outskirts.

astro-ph.GA↗

arXiv2022

Self-Taught Metric Learning without Labels

We present a novel self-taught framework for unsupervised metric learning, which alternates between predicting class-equivalence relations between data through a moving average of an embedding model and learning the model with the predicted relations as pseudo labels. At the heart of our framework lies an algorithm that investigates contexts of data on the embedding space to predict their class-equivalence relations as pseudo labels. The algorithm enables efficient end-to-end training since it demands no off-the-shelf module for pseudo labeling. Also, the class-equivalence relations provide rich supervisory signals for learning an embedding space. On standard benchmarks for metric learning, it clearly outperforms existing unsupervised learning methods and sometimes even beats supervised learning models using the same backbone network. It is also applied to semi-supervised metric learning as a way of exploiting additional unlabeled data, and achieves the state of the art by boosting performance of supervised learning substantially.

cs.CV↗

arXiv2022

ReSTR: Convolution-free Referring Image Segmentation Using Transformers

Referring image segmentation is an advanced semantic segmentation task where target is not a predefined class but is described in natural language. Most of existing methods for this task rely heavily on convolutional neural networks, which however have trouble capturing long-range dependencies between entities in the language expression and are not flexible enough for modeling interactions between the two different modalities. To address these issues, we present the first convolution-free model for referring image segmentation using transformers, dubbed ReSTR. Since it extracts features of both modalities through transformer encoders, it can capture long-range dependencies between entities within each modality. Also, ReSTR fuses features of the two modalities by a self-attention encoder, which enables flexible and adaptive interactions between the two modalities in the fusion process. The fused features are fed to a segmentation module, which works adaptively according to the image and language expression in hand. ReSTR is evaluated and compared with previous work on all public benchmarks, where it outperforms all existing models.

cs.CV↗

arXiv2021

The 3-D Kinematics of the Orion Nebula Cluster: NIRSPEC-AO Radial Velocities of the Core Population

The kinematics and dynamics of stellar and substellar populations within young, still-forming clusters provides valuable information for constraining theories of formation mechanisms. Using Keck II NIRSPEC+AO data, we have measured radial velocities for 56 low-mass sources within 4' of the core of the Orion Nebula Cluster (ONC). We also re-measure radial velocities for 172 sources observed with SDSS/APOGEE. These data are combined with proper motions measured using $HST$ ACS/WFPC2/WFC3IR and Keck II NIRC2, creating a sample of 135 sources with all three velocity components. The velocities measured are consistent with a normal distribution in all three components. We measure intrinsic velocity dispersions of ($σ_{v_α}$, $σ_{v_δ}$, $σ_{v_r}$) = ($1.64\pm0.12$, $2.03\pm0.13$, $2.56^{+0.16}_{-0.17}$) km s$^{-1}$. Our computed intrinsic velocity dispersion profiles are consistent with the dynamical equilibrium models from Da Rio et al. (2014) in the tangential direction, but not in the line of sight direction, possibly indicating that the core of the ONC is not yet virialized, and may require a non-spherical potential to explain the observed velocity dispersion profiles. We also observe a slight elongation along the north-south direction following the filament, which has been well studied in previous literature, and an elongation in the line of sight to tangential velocity direction. These 3-D kinematics will help in the development of realistic models of the formation and early evolution of massive clusters.

astro-ph.GA↗

arXiv2021

Embedding Transfer with Label Relaxation for Improved Metric Learning

This paper presents a novel method for embedding transfer, a task of transferring knowledge of a learned embedding model to another. Our method exploits pairwise similarities between samples in the source embedding space as the knowledge, and transfers them through a loss used for learning target embedding models. To this end, we design a new loss called relaxed contrastive loss, which employs the pairwise similarities as relaxed labels for inter-sample relations. Our loss provides a rich supervisory signal beyond class equivalence, enables more important pairs to contribute more to training, and imposes no restriction on manifolds of target embedding spaces. Experiments on metric learning benchmarks demonstrate that our method largely improves performance, or reduces sizes and output dimensions of target models effectively. We further show that it can be also used to enhance quality of self-supervised representation and performance of classification models. In all the experiments, our method clearly outperforms existing embedding transfer techniques.

cs.CV↗

arXiv2021

An extended halo around an ancient dwarf galaxy

The Milky Way is surrounded by dozens of ultra-faint (< $10^5$ solar luminosities) dwarf satellite galaxies. They are the surviving remnants of the earliest galaxies, as confirmed by their ancient (~13 billion years old) and chemically primitive stars. Simulations suggest that these systems formed within extended dark matter halos and experienced early galaxy mergers and supernova feedback. However, the signatures of these events would lie outside their core regions (>2 half-light radii), which are spectroscopically unstudied due to the sparseness of their distant stars. Here we identify members of the Tucana II ultra-faint dwarf galaxy in its outer region (up to 9 half-light radii), demonstrating the system to be dramatically more spatially extended and chemically primitive than previously found. These distant stars are extremely metal-poor (<[Fe/H]>=-3.02; less than ~1/1000th of the solar iron abundance), affirming Tucana II as the most metal-poor known galaxy. We observationally establish, for the first time, an extended dark matter halo surrounding an ultra-faint dwarf galaxy out to one kiloparsec, with a total mass of >$10^7$ solar masses. This measurement is consistent with the expected ~2x$10^7$ solar masses using a generalized NFW density profile. The extended nature of Tucana II suggests that it may have undergone strong bursty feedback or been the product of an early galactic merger. We demonstrate that spatially extended stellar populations, which other ultra-faint dwarfs hint at hosting as well, are observable in principle and open the possibility for detailed studies of the stellar halos of relic galaxies.

astro-ph.GA↗

arXiv2020

Proxy Anchor Loss for Deep Metric Learning

Existing metric learning losses can be categorized into two classes: pair-based and proxy-based losses. The former class can leverage fine-grained semantic relations between data points, but slows convergence in general due to its high training complexity. In contrast, the latter class enables fast and reliable convergence, but cannot consider the rich data-to-data relations. This paper presents a new proxy-based loss that takes advantages of both pair- and proxy-based methods and overcomes their limitations. Thanks to the use of proxies, our loss boosts the speed of convergence and is robust against noisy labels and outliers. At the same time, it allows embedding vectors of data to interact with each other in its gradients to exploit data-to-data relations. Our method is evaluated on four public benchmarks, where a standard network trained with our loss achieves state-of-the-art performance and most quickly converges.

cs.CV↗

arXiv2020

Stellar metallicities from SkyMapper photometry I: A study of the Tucana II ultra-faint dwarf galaxy

We present a study of the ultra-faint Milky Way dwarf satellite galaxy Tucana II using deep photometry from the 1.3m SkyMapper telescope at Siding Spring Observatory, Australia. The SkyMapper filter-set contains a metallicity-sensitive intermediate-band $v$ filter covering the prominent Ca II K feature at 3933.7A. When combined with photometry from the SkyMapper $u, g$, and $i$ filters, we demonstrate that $v$ band photometry can be used to obtain stellar metallicities with a precision of $\sim0.20$dex when [Fe/H] $> -2.5$, and $\sim0.34$dex when [Fe/H] $< -2.5$. Since the $u$ and $v$ filters bracket the Balmer Jump at 3646A, we also find that the filter-set can be used to derive surface gravities. We thus derive photometric metallicities and surface gravities for all stars down to a magnitude of $g\sim20$ within $\sim$75 arcminutes of Tucana II. Photometric metallicity and surface gravity cuts remove nearly all foreground contamination. By incorporating Gaia proper motions, we derive quantitative membership probabilities which recover all known members on the red giant branch of Tucana II. Additionally, we identify multiple likely new members in the center of the system and candidate members several half-light radii from the center of the system. Finally, we present a metallicity distribution function derived from the photometric metallicities of likely Tucana II members. This result demonstrates the utility of wide-field imaging with the SkyMapper filter-set in studying UFDs, and in general, low surface brightness populations of metal-poor stars. Upcoming work will clarify the membership status of several distant stars identified as candidate members of Tucana II.

astro-ph.GA↗

arXiv2019

RR Lyrae Stars in the Field of Sagittarius II

We present the detection of RR Lyrae variable stars in the field of the Sagittarius II (Sgr II) ultra-faint dwarf (UFD) galaxy. Using B, V time-series photometry obtained with the Korea Microlensing Telescope Network (KMTNet) 1.6 m telescope at CTIO and G-band data from Gaia Data Release 2 (DR2), we identified and characterized two ab-type and four c-type RR Lyrae variables. Five out of the six stars are clustered within three half-light radii (~4.8') of the galaxy indicating their association with Sgr II, while the RRab star V4 is located ~22' from the galaxy center. By excluding V4, the high c-type fraction (0.8) and the period of the only RRab star V3 (P_V3 = 0.666 days) suggest an Oosterhoff II (Oo II) classification for Sgr II. Located close to the locus of Oo II clusters in the period-amplitude diagram, V3 is similar to RRab stars in other UFDs having Oosterhoff-intermediate and Oo II properties. Sgr II is, however, more compact than usual UFDs, placed in between star clusters and dwarf galaxies in the size-luminosity plane, and therefore spectroscopic studies are eventually required to ascertain the true nature of this stellar system. We derive the metallicity ([Fe/H]_RRab = -2.1 +- 0.3) and heliocentric distance (~64 +- 3 kpc) of Sgr II from the RR Lyrae stars, and estimate its age (~12 Gyr) based on our stellar population models. The Oosterhoff properties of UFDs can be explained with the evolution effect of RR Lyrae stars in the instability strip.

astro-ph.GA↗