arXiv subjects

Sangin Lee

Publications and source records attributed to Sangin Lee.

4 recordsLinked to original sources

arXiv2026

Multi-Modal Guided Multi-Source Domain Adaptation for Object Detection

General object detection (OD) struggles to detect objects in the target domain that differ from the training distribution. To address this, recent studies demonstrate that training from multiple source domains and explicitly processing them separately for multi-source domain adaptation (MSDA) outperforms blending them for unsupervised domain adaptation (UDA). However, existing MSDA methods learn domain-agnostic features from domain-specific RGB images while preserving domain-specific information from the domain-agnostic feature map. To address this, we propose MS-DePro: Multi-Source Detector with Depth and Prompt, composed of (1) depth-guided localization and (2) multi-modal guided prompt learning. We leverage domain-agnostic input modalities, namely depth maps and text, to encode domain-agnostic characteristics. Specifically, we utilize depth maps to generate domain-agnostic region proposals for localization and integrate multi-modal features to align learnable text embeddings for classification. MS-DePro achieves state-of-the-art performance on MSDA benchmarks, and comprehensive ablations demonstrate the effectiveness of our contributions. Our code is available on https://github.com/sejong-rcv/Multi-Modal-Guided-Multi-Source-Domain-Adaptation-for-Object-Detection.

cs.CV↗

arXiv2026

CLIP Tricks You: Training-free Token Pruning for Efficient Pixel Grounding in Large VIsion-Language Models

In large vision-language models, visual tokens typically constitute the majority of input tokens, leading to substantial computational overhead. To address this, recent studies have explored pruning redundant or less informative visual tokens for image understanding tasks. However, these methods struggle with pixel grounding tasks, where token importance is highly contingent on the input text. Through an in-depth analysis of CLIP, we observe that visual tokens within referent regions often exhibit low similarity to their textual representation. Motivated by this insight, we introduce LiteLVLM, a training-free, text-guided token pruning strategy for efficient pixel grounding inference. By reversing the ranking of CLIP's visual-text similarity, LiteLVLM effectively retains visual tokens covering the referent regions, while recovering context tokens to enable clear foreground-background separation. Extensive experiments demonstrate that LiteLVLM significantly outperforms existing methods by over 5% across diverse token budgets. Without any training or fine-tuning, LiteLVLM maintains 90% of the original performance with a 22% speedup and a 2.3X memory reduction. Our code is available at https://github.com/sejong-rcv/LiteLVLM.

cs.CV↗

arXiv2018

A unified algorithm for the non-convex penalized estimation: The ncpen package

Various R packages have been developed for the non-convex penalized estimation but they can only be applied to the smoothly clipped absolute deviation (SCAD) or minimax concave penalty (MCP). We develop an R package, entitled ncpen, for the non-convex penalized estimation in order to make data analysts to experience other non-convex penalties. The package ncpen implements a unified algorithm based on the convex concave procedure and modified local quadratic approximation algorithm, which can be applied to a broader range of non-convex penalties, including the SCAD and MCP as special examples. Many user-friendly functionalities such as generalized information criteria, cross-validation and L2-stabilization are provided also.

stat.CO↗

arXiv2014

Strong rules for nonconvex penalties and their implications for efficient algorithms in high-dimensional regression

We consider approaches for improving the efficiency of algorithms for fitting nonconvex penalized regression models such as SCAD and MCP in high dimensions. In particular, we develop rules for discarding variables during cyclic coordinate descent. This dimension reduction leads to a substantial improvement in the speed of these algorithms for high-dimensional problems. The rules we propose here eliminate a substantial fraction of the variables from the coordinate descent algorithm. Violations are quite rare, especially in the locally convex region of the solution path, and furthermore, may be easily detected and corrected by checking the Karush-Kuhn-Tucker conditions. We extend these rules to generalized linear models, as well as to other nonconvex penalties such as the $\ell_2$-stabilized Mnet penalty, group MCP, and group SCAD. We explore three variants of the coordinate decent algorithm that incorporate these rules and study the efficiency of these algorithms in fitting models to both simulated data and on real data from a genome-wide association study.

stat.CO↗