A5061636336- Advanced Neural Network Applications
- Advanced Image and Video Retrieval Techniques
- Rare-earth and actinide compounds
- Multimodal Machine Learning Applications
- Topic Modeling
- Human Pose and Action Recognition
- Natural Language Processing Techniques
- Iron-based superconductors research
- Generative Adversarial Networks and Image Synthesis
- Video Surveillance and Tracking Methods
- Magnetic and transport properties of perovskites and related materials
- Advanced Vision and Imaging
- High Entropy Alloys Studies
- Image Enhancement Techniques
- Domain Adaptation and Few-Shot Learning
- Image Retrieval and Classification Techniques
- Advanced materials and composites
- Supercapacitor Materials and Fabrication
- Polymer Surface Interaction Studies
- Nanocomposite Films for Food Packaging
- Inorganic Chemistry and Materials
- Visual Attention and Saliency Detection
- Advancements in Battery Materials
- High-Temperature Coating Behaviors
- Video Analysis and Summarization
Samsung (South Korea)
2023-2025
Korea University
2021-2022
Sejong University
2022
Sungkyunkwan University
2002-2021
Suwon Research Institute
2019-2020
Government of the Republic of Korea
2019-2020
Tokushima University
1991
Achieving low thermal conductivity and high mechanical strength presents a material design challenge due to intrinsic trade-offs, such as the aerogel's porosity, impeding applications in construction, industry, aerospace. This study composite that incorporates silica aerogel within thermally expanded 2D layered vermiculite matrix. overcomes limitations imposed by van der Waals bonding lengths, typically less than 10 Å, which hinder integration with crystals. Our method employs spark reaction...
Polymeric three-dimensional inverse-opal (IO) structures provide unique structural properties useful for various applications ranging from optics to separation technologies. Despite vast needs IO functionalization impart additional chemical properties, this task has been seriously challenged by the intrinsic limitation of polymeric porous materials that do not allow easy penetration waterborne moieties or precursors. To overcome restriction, we present a robust and straightforward method...
We propose a novel cost aggregation network, called Cost Aggregation Transformers (CATs), to find dense correspondences between semantically similar images with additional challenges posed by large intra-class appearance and geometric variations. is highly important process in matching tasks, which the accuracy depends on quality of its output. Compared hand-crafted or CNN-based methods addressing aggregation, that either lacks robustness severe deformations inherit limitation CNNs fail...
Zwitterionic sulfobetaine moieties were anchored on a separator surface to concurrently facilitate selective Li-ion transport and polysulfide conversion, finally resulting in significantly enhanced performance for durable Li–S battery operation.
With growing concerns over electronic device malfunction and the resulting information loss caused by electromagnetic interference (EMI), extensive studies have been performed in developing EMI shielding techniques.
The proteolytic activities of the 20 S proteasome were found to change in their levels during development chick embryonic muscle.The peptide-cleaving against N-succinyl-Leu-Leu-Val-Tyr-7amido-4-methylcoumarin and N-benzyloxycarbonyl-Ala-Arg-Arg-4-methoxy-&naphthylamide gradually decreased with time development.On other hand, casein-degrading activity presence poly-L-lysine markedly increased from day 11 reached a maximal level by 17.These changes appeared be tissue-specific because little or...
Binder-free layer-by-layer assembled multilayers consisting of reduced graphene oxide and alumina nanoparticles are prepared for implementing heat dissipation films with outstandingly high in-plane cross-plane thermal conductivities.
Cost aggregation is a process in image matching tasks that aims to disambiguate the noisy scores. Existing methods generally tackle this by hand-crafted or CNN-based methods, which either lack robustness severe deformations inherit limitation of CNNs fail discriminate incorrect matches due limited receptive fields and inadaptability. In paper, we introduce Aggregation with Transformers (CATs) exploring global consensus among initial correlation map help some architectural designs allow us...
In the paradigm of AI-generated content (AIGC), there has been increasing attention to transferring knowledge from pre-trained text-to-image (T2I) models text-to-video (T2V) generation. Despite their effectiveness, these frameworks face challenges in maintaining consistent narratives and handling shifts scene composition or object placement a single abstract user prompt. Exploring ability large language (LLMs) generate time-dependent, frame-by-frame prompts, this paper introduces new...
Conventional techniques to establish dense correspondences across visually or semantically similar images focused on designing a task-specific matching prior, which is difficult model in general. To overcome this, recent learning-based methods have attempted learn good prior within itself large training data. The performance improvement was apparent, but the need for sufficient data and intensive learning hinders their applicability. Moreover, using fixed at test time does not account fact...
We introduce a novel cost aggregation network, dubbed Volumetric Aggregation with Transformers (VAT), to tackle the few-shot segmentation task by using both convolutions and transformers efficiently handle high dimensional correlation maps between query support. In specific, we propose our encoder consisting of volume embedding module not only transform into more tractable size but also inject some convolutional inductive bias volumetric transformer for aggregation. Our has pyramidal...
We present a novel architecture for dense correspondence. The current state-of-the-art are Transformer-based approaches that focus on either feature descriptors or cost volume aggregation. However, they generally aggregate one the other but not both, though joint aggregation would boost each by providing information has lacks, i.e., structural semantic of an image, pixel-wise matching similarity. In this work, we propose network interleaves both forms aggregations in way exploits their...
Open-vocabulary semantic segmentation presents the challenge of labeling each pixel within an image based on a wide range text descriptions. In this work, we introduce novel cost-based approach to adapt vision-language foundation models, notably CLIP, for intricate task segmentation. Through aggregating cosine similarity score, i.e., cost volume between and embeddings, our method potently adapts CLIP segmenting seen unseen classes by fine-tuning its encoders, addressing challenges faced...
Successful sequential recommendation systems rely on accurately capturing the user's short-term and long-term interest. Although Transformer-based models achieved state-of-the-art performance in task, they generally require quadratic memory time complexity to sequence length, making it difficult extract interest of users. On other hand, Multi-Layer Perceptrons (MLP)-based models, renowned for their linear complexity, have recently shown competitive results compared Transformer various tasks....
This paper introduces a Transformer-based integrative feature and cost aggregation network designed for dense matching tasks. In the context of matching, many works benefit from one two forms aggregation: aggregation, which pertains to alignment similar features, or procedure aimed at instilling coherence in flow estimates across neighboring pixels. this work, we first show that exhibit distinct characteristics reveal potential substantial benefits stemming judicious use both processes. We...
Large-scale vision-language models like CLIP have demonstrated impressive open-vocabulary capabilities for image-level tasks, excelling in recognizing what objects are present. However, they struggle with pixel-level recognition tasks semantic segmentation, which additionally require understanding where the located. In this work, we propose a novel method, PixelCLIP, to adapt image encoder by guiding model on where, is achieved using unlabeled images and masks generated from vision...
In this work, we explore new perspectives on cross-view completion learning by drawing an analogy to self-supervised correspondence learning. Through our analysis, demonstrate that the cross-attention map within models captures more effectively than other correlations derived from encoder or decoder features. We verify effectiveness of evaluating both zero-shot matching and learning-based geometric multi-frame depth estimation. Project page is available at https://cvlab-kaist.github.io/ZeroCo/.