Sang Hyun Park

Sang Hyun Park, Jun Young Koh, Junha Lee et al.

In this work, we share the insights for achieving state-of-the-art quality in our text-to-image anime image generative model, called Illustrious. To achieve high resolution, dynamic color range images, and high restoration ability, we focus on three critical approaches for model improvement. First, we delve into the significance of the batch size and dropout control, which enables faster learning of controllable token based concept activations. Second, we increase the training resolution of images, affecting the accurate depiction of character anatomy in much higher resolution, extending its generation capability over 20MP with proper methods. Finally, we propose the refined multi-level captions, covering all tags and various natural language captions as a critical factor for model development. Through extensive analysis and experiments, Illustrious demonstrates state-of-the-art performance in terms of animation style, outperforming widely-used models in illustration domains, propelling easier customization and personalization with nature of open source. We plan to publicly release updated Illustrious model series sequentially as well as sustainable plans for improvements.

Philip Chikontwe, Soo Jeong Nam, Heounjeong Go et al.

Whole slide image (WSI) classification is a fundamental task for the diagnosis and treatment of diseases; but, curation of accurate labels is time-consuming and limits the application of fully-supervised methods. To address this, multiple instance learning (MIL) is a popular method that poses classification as a weakly supervised learning task with slide-level labels only. While current MIL methods apply variants of the attention mechanism to re-weight instance features with stronger models, scant attention is paid to the properties of the data distribution. In this work, we propose to re-calibrate the distribution of a WSI bag (instances) by using the statistics of the max-instance (critical) feature. We assume that in binary MIL, positive bags have larger feature magnitudes than negatives, thus we can enforce the model to maximize the discrepancy between bags with a metric feature loss that models positive bags as out-of-distribution. To achieve this, unlike existing MIL methods that use single-batch training modes, we propose balanced-batch sampling to effectively use the feature loss i.e., (+/-) bags simultaneously. Further, we employ a position encoding module (PEM) to model spatial/morphological information, and perform pooling by multi-head self-attention (PSMA) with a Transformer encoder. Experimental results on existing benchmark datasets show our approach is effective and improves over state-of-the-art MIL methods.

Philip Chikontwe, Soopil Kim, Sang Hyun Park

Few-shot classification is a challenging problem that aims to learn a model that can adapt to unseen classes given a few labeled samples. Recent approaches pre-train a feature extractor, and then fine-tune for episodic meta-learning. Other methods leverage spatial features to learn pixel-level correspondence while jointly training a classifier. However, results using such approaches show marginal improvements. In this paper, inspired by the transformer style self-attention mechanism, we propose a strategy to cross-attend and re-weight discriminative features for few-shot classification. Given a base representation of support and query images after global pooling, we introduce a single shared module that projects features and cross-attends in two aspects: (i) query to support, and (ii) support to query. The module computes attention scores between features to produce an attention pooled representation of features in the same class that is later added to the original representation followed by a projection head. This effectively re-weights features in both aspects (i & ii) to produce features that better facilitate improved metric-based meta-learning. Extensive experiments on public benchmarks show our approach outperforms state-of-the-art methods by 3%~5%.

Myeongkyun Kang, Soopil Kim, Xiaoxiao Li et al.

Large vision language models (LVLMs) have demonstrated impressive performance across a wide range of tasks. These capabilities largely stem from visual instruction tuning, which fine-tunes models on datasets consisting of curated image-instruction-output triplets. However, in the medical domain, constructing large-scale, high-quality instruction datasets is particularly challenging due to the need for specialized expert knowledge. To address this issue, we propose an instruction-free tuning approach that reduces reliance on handcrafted instructions, leveraging only image-description pairs for fine-tuning. Specifically, we introduce a momentum proxy instruction as a replacement for curated text instructions, which preserves the instruction-following capability of the pre-trained LVLM while promoting updates to parameters that remain valid during inference. Consequently, the fine-tuned LVLM can flexibly respond to domain-specific instructions, even though explicit instructions are absent during fine-tuning. Additionally, we incorporate a response shuffling strategy to mitigate the model's over-reliance on previous words, facilitating more effective fine-tuning. Our approach achieves state-of-the-art accuracy on multiple-choice visual question answering tasks across SKINCON, WBCAtt, CBIS, and MIMIC-CXR datasets, significantly enhancing the fine-tuning efficiency of LVLMs in medical domains.

Sion An, Myeongkyun Kang, Soopil Kim et al.

Electroencephalography (EEG) motor imagery (MI) classification is a fundamental, yet challenging task due to the variation of signals between individuals i.e., inter-subject variability. Previous approaches try to mitigate this using task-specific (TS) EEG signals from the target subject in training. However, recording TS EEG signals requires time and limits its applicability in various fields. In contrast, resting state (RS) EEG signals are a viable alternative due to ease of acquisition with rich subject information. In this paper, we propose a novel subject-adaptive transfer learning strategy that utilizes RS EEG signals to adapt models on unseen subject data. Specifically, we disentangle extracted features into task- and subject-dependent features and use them to calibrate RS EEG signals for obtaining task information while preserving subject characteristics. The calibrated signals are then used to adapt the model to the target subject, enabling the model to simulate processing TS EEG signals of the target subject. The proposed method achieves state-of-the-art accuracy on three public benchmarks, demonstrating the effectiveness of our method in cross-subject EEG MI classification. Our findings highlight the potential of leveraging RS EEG signals to advance practical brain-computer interface systems. The code is available at https://github.com/SionAn/MICCAI2024-ResTL.

Seunghun Lee, Jiwan Seo, Jeonghoon Kim et al.

Referring Video Object Segmentation (RVOS) aims to segment and track objects in videos based on natural language expressions, requiring precise alignment between visual content and textual queries. However, existing methods often suffer from semantic misalignment, largely due to indiscriminate frame sampling and supervision of all visible objects during training -- regardless of their actual relevance to the expression. We identify the core problem as the absence of an explicit temporal learning signal in conventional training paradigms. To address this, we introduce MeViS-M, a dataset built upon the challenging MeViS benchmark, where we manually annotate temporal spans when each object is referred to by the expression. These annotations provide a direct, semantically grounded supervision signal that was previously missing. To leverage this signal, we propose Temporally Grounded Learning (TGL), a novel learning framework that directly incorporates temporal grounding into the training process. Within this frame- work, we introduce two key strategies. First, Moment-guided Dual-path Propagation (MDP) improves both grounding and tracking by decoupling language-guided segmentation for relevant moments from language-agnostic propagation for others. Second, Object-level Selective Supervision (OSS) supervises only the objects temporally aligned with the expression in each training clip, thereby reducing semantic noise and reinforcing language-conditioned learning. Extensive experiments demonstrate that our TGL framework effectively leverages temporal signal to establish a new state-of-the-art on the challenging MeViS benchmark. We will make our code and the MeViS-M dataset publicly available.

Woo Kyoung Han, Yongjun Lee, Byeonghun Lee et al.

Despite significant advances in learning-based lossy compression algorithms, standardizing codecs remains a critical challenge. In this paper, we present the JPEG Processing Neural Operator (JPNeO), a next-generation JPEG algorithm that maintains full backward compatibility with the current JPEG format. Our JPNeO improves chroma component preservation and enhances reconstruction fidelity compared to existing artifact removal methods by incorporating neural operators in both the encoding and decoding stages. JPNeO achieves practical benefits in terms of reduced memory usage and parameter count. We further validate our hypothesis about the existence of a space with high mutual information through empirical evidence. In summary, the JPNeO functions as a high-performance out-of-the-box image compression pipeline without changing source coding's protocol. Our source code is available at https://github.com/WooKyoungHan/JPNeO.

Soopil Kim, Sion An, Philip Chikontwe et al.

Logical anomalies (LA) refer to data violating underlying logical constraints e.g., the quantity, arrangement, or composition of components within an image. Detecting accurately such anomalies requires models to reason about various component types through segmentation. However, curation of pixel-level annotations for semantic segmentation is both time-consuming and expensive. Although there are some prior few-shot or unsupervised co-part segmentation algorithms, they often fail on images with industrial object. These images have components with similar textures and shapes, and a precise differentiation proves challenging. In this study, we introduce a novel component segmentation model for LA detection that leverages a few labeled samples and unlabeled images sharing logical constraints. To ensure consistent segmentation across unlabeled images, we employ a histogram matching loss in conjunction with an entropy loss. As segmentation predictions play a crucial role, we propose to enhance both local and global sample validity detection by capturing key aspects from visual semantics via three memory banks: class histograms, component composition embeddings and patch-level representations. For effective LA detection, we propose an adaptive scaling strategy to standardize anomaly scores from different memory banks in inference. Extensive experiments on the public benchmark MVTec LOCO AD reveal our method achieves 98.1% AUROC in LA detection vs. 89.6% from competing methods.

Yunseong Nam, Jiwon Jang, Dongkyu Won et al.

Preference optimization offers a scalable supervision paradigm based on relative preference signals, yet prior attempts in medical image segmentation remain model-specific and rely on low-diversity prediction sampling. In this paper, we propose MAPO (Model-Agnostic Preference Optimization), a training framework that utilizes Dropout-driven stochastic segmentation hypotheses to construct preference-consistent gradients without direct ground-truth supervision. MAPO is fully architecture- and dimensionality-agnostic, supporting 2D/3D CNN and Transformer-based segmentation pipelines. Comprehensive evaluations across diverse medical datasets reveal that MAPO consistently enhances boundary adherence, reduces overfitting, and yields more stable optimization dynamics compared to conventional supervised training.

Jae Wan Park, Sang Hyun Park, Jun Young Koh et al.

The emergence of various adapters, including Low-Rank Adaptation (LoRA) applied from the field of natural language processing, has allowed diffusion models to personalize image generation at a low cost. However, due to the various challenges including limited datasets and shortage of regularization and computation resources, adapter training often results in unsatisfactory outcomes, leading to the corruption of the backbone model's prior knowledge. One of the well known phenomena is the loss of diversity in object generation, especially within the same class which leads to generating almost identical objects with minor variations. This poses challenges in generation capabilities. To solve this issue, we present Contrastive Adapter Training (CAT), a simple yet effective strategy to enhance adapter training through the application of CAT loss. Our approach facilitates the preservation of the base model's original knowledge when the model initiates adapters. Furthermore, we introduce the Knowledge Preservation Score (KPS) to evaluate CAT's ability to keep the former information. We qualitatively and quantitatively compare CAT's improvement. Finally, we mention the possibility of CAT in the aspects of multi-concept adapter and optimization.

Seunghun Lee, Jiwan Seo, Minwoo Choi et al.

In this paper, we present Latest Object Memory Management (LOMM) for temporally consistent video instance segmentation that significantly improves long-term instance tracking. At the core of our method is Latest Object Memory (LOM), which robustly tracks and continuously updates the latest states of objects by explicitly modeling their presence in each frame. This enables consistent tracking and accurate identity management across frames, enhancing both performance and reliability through the VIS process. Moreover, we introduce Decoupled Object Association (DOA), a strategy that separately handles newly appearing and already existing objects. By leveraging our memory system, DOA accurately assigns object indices, improving matching accuracy and ensuring stable identity consistency, even in dynamic scenes where objects frequently appear and disappear. Extensive experiments and ablation studies demonstrate the superiority of our method over traditional approaches, setting a new benchmark in VIS. Notably, our LOMM achieves state-of-the-art AP score of 54.0 on YouTube-VIS 2022, a dataset known for its challenging long videos. Project page: https://seung-hun-lee.github.io/projects/LOMM/

Jun Young Koh, Sang Hyun Park, Joy Song

The recent emergence of latent diffusion models such as SDXL and SD 1.5 has shown significant capability in generating highly detailed and realistic images. Despite their remarkable ability to produce images, generating accurate text within images still remains a challenging task. In this paper, we examine the validity of fine-tuning approaches in generating legible text within the image. We propose a low-cost approach by leveraging SDXL without any time-consuming training on large-scale datasets. The proposed strategy employs a fine-tuning technique that examines the effects of data refinement levels and synthetic captions. Moreover, our results demonstrate how our small scale fine-tuning approach can improve the accuracy of text generation in different scenarios without the need of additional multimodal encoders. Our experiments show that with the addition of random letters to our raw dataset, our model's performance improves in producing well-formed visual text.

Soopil Kim, Philip Chikontwe, Sang Hyun Park

Few shot segmentation (FSS) aims to learn pixel-level classification of a target object in a query image using only a few annotated support samples. This is challenging as it requires modeling appearance variations of target objects and the diverse visual cues between query and support images with limited information. To address this problem, we propose a semi-supervised FSS strategy that leverages additional prototypes from unlabeled images with uncertainty guided pseudo label refinement. To obtain reliable prototypes from unlabeled images, we meta-train a neural network to jointly predict segmentation and estimate the uncertainty of predictions. We employ the uncertainty estimates to exclude predictions with high degrees of uncertainty for pseudo label construction to obtain additional prototypes based on the refined pseudo labels. During inference, query segmentation is predicted using prototypes from both support and unlabeled images including low-level features of the query images. Our approach is end-to-end and can easily supplement existing approaches without the requirement of additional training to employ unlabeled samples. Extensive experiments on PASCAL-$5^i$ and COCO-$20^i$ demonstrate that our model can effectively remove unreliable predictions to refine pseudo labels and significantly improve upon state-of-the-art performances.

Myeongkyun Kang, Dongkyu Won, Miguel Luna et al.

Models trained on datasets with texture bias usually perform poorly on out-of-distribution samples since biased representations are embedded into the model. Recently, various image translation and debiasing methods have attempted to disentangle texture biased representations for downstream tasks, but accurately discarding biased features without altering other relevant information is still challenging. In this paper, we propose a novel framework that leverages image translation to generate additional training images using the content of a source image and the texture of a target image with a different bias property to explicitly mitigate texture bias when training a model on a target task. Our model ensures texture similarity between the target and generated images via a texture co-occurrence loss while preserving content details from source images with a spatial self-similarity loss. Both the generated and original training images are combined to train improved classification or segmentation models robust to inconsistent texture bias. Evaluation on five classification- and two segmentation-datasets with known texture biases demonstrates the utility of our method, and reports significant improvements over recent state-of-the-art methods in all cases.

Heejung Park, Gyeong Min Lee, Soopil Kim et al.

Non-rigid registration is a necessary but challenging task in medical imaging studies. Recently, unsupervised registration models have shown good performance, but they often require a large-scale training dataset and long training times. Therefore, in real world application where only dozens to hundreds of image pairs are available, existing models cannot be practically used. To address these limitations, we propose a novel unsupervised registration model which is integrated with a gradient-based meta learning framework. In particular, we train a meta learner which finds an initialization point of parameters by utilizing a variety of existing registration datasets. To quickly adapt to various tasks, the meta learner was updated to get close to the center of parameters which are fine-tuned for each registration task. Thereby, our model can adapt to unseen domain tasks via a short fine-tuning process and perform accurate registration. To verify the superiority of our model, we train the model for various 2D medical image registration tasks such as retinal choroid Optical Coherence Tomography Angiography (OCTA), CT organs, and brain MRI scans and test on registration of retinal OCTA Superficial Capillary Plexus (SCP). In our experiments, the proposed model obtained significantly improved performance in terms of accuracy and training time compared to other registration models.

Dongkyu Won, Euijin Jung, Sion An et al.

Recently, Self-supervised learning methods able to perform image denoising without ground truth labels have been proposed. These methods create low-quality images by adding random or Gaussian noise to images and then train a model for denoising. Ideally, it would be beneficial if one can generate high-quality CT images with only a few training samples via self-supervision. However, the performance of CT denoising is generally limited due to the complexity of CT noise. To address this problem, we propose a novel self-supervised learning-based CT denoising method. In particular, we train pre-train CT denoising and noise models that can predict CT noise from Low-dose CT (LDCT) using available LDCT and Normal-dose CT (NDCT) pairs. For a given test LDCT, we generate Pseudo-LDCT and NDCT pairs using the pre-trained denoising and noise models and then update the parameters of the denoising model using these pairs to remove noise in the test LDCT. To make realistic Pseudo LDCT, we train multiple noise models from individual images and generate the noise using the ensemble of noise models. We evaluate our method on the 2016 AAPM Low-Dose CT Grand Challenge dataset. The proposed ensemble noise model can generate realistic CT noise, and thus our method significantly improves the denoising performance existing denoising models trained by supervised- and self-supervised learning.

Myeongkyun Kang, Philip Chikontwe, Miguel Luna et al.

Following the pandemic outbreak, several works have proposed to diagnose COVID-19 with deep learning in computed tomography (CT); reporting performance on-par with experts. However, models trained/tested on the same in-distribution data may rely on the inherent data biases for successful prediction, failing to generalize on out-of-distribution samples or CT with different scanning protocols. Early attempts have partly addressed bias-mitigation and generalization through augmentation or re-sampling, but are still limited by collection costs and the difficulty of quantifying bias in medical images. In this work, we propose Mixing-AdaSIN; a bias mitigation method that uses a generative model to generate de-biased images by mixing texture information between different labeled CT scans with semantically similar features. Here, we use Adaptive Structural Instance Normalization (AdaSIN) to enhance de-biasing generation quality and guarantee structural consistency. Following, a classifier trained with the generated images learns to correctly predict the label without bias and generalizes better. To demonstrate the efficacy of our method, we construct a biased COVID-19 vs. bacterial pneumonia dataset based on CT protocols and compare with existing state-of-the-art de-biasing methods. Our experiments show that classifiers trained with de-biased generated images report improved in-distribution performance and generalization on an external COVID-19 dataset.

Soopil Kim, Sion An, Philip Chikontwe et al.

Segmentation of organs of interest in 3D medical images is necessary for accurate diagnosis and longitudinal studies. Though recent advances using deep learning have shown success for many segmentation tasks, large datasets are required for high performance and the annotation process is both time consuming and labor intensive. In this paper, we propose a 3D few shot segmentation framework for accurate organ segmentation using limited training samples of the target organ annotation. To achieve this, a U-Net like network is designed to predict segmentation by learning the relationship between 2D slices of support data and a query image, including a bidirectional gated recurrent unit (GRU) that learns consistency of encoded features between adjacent slices. Also, we introduce a transfer learning method to adapt the characteristics of the target image and organ by updating the model before testing with arbitrary support and query data sampled from the support data. We evaluate our proposed model using three 3D CT datasets with annotations of different organs. Our model yielded significantly improved performance over state-of-the-art few shot segmentation models and was comparable to a fully supervised model trained with more target training data.

Sion An, Soopil Kim, Philip Chikontwe et al.

Brain-Computer Interfaces (BCI) based on Electroencephalography (EEG) signals, in particular motor imagery (MI) data have received a lot of attention and show the potential towards the design of key technologies both in healthcare and other industries. MI data is generated when a subject imagines movement of limbs and can be used to aid rehabilitation as well as in autonomous driving scenarios. Thus, classification of MI signals is vital for EEG-based BCI systems. Recently, MI EEG classification techniques using deep learning have shown improved performance over conventional techniques. However, due to inter-subject variability, the scarcity of unseen subject data, and low signal-to-noise ratio, extracting robust features and improving accuracy is still challenging. In this context, we propose a novel two-way few shot network that is able to efficiently learn how to learn representative features of unseen subject categories and how to classify them with limited MI EEG data. The pipeline includes an embedding module that learns feature representations from a set of samples, an attention mechanism for key signal feature discovery, and a relation module for final classification based on relation scores between a support set and a query signal. In addition to the unified learning of feature similarity and a few shot classifier, our method leads to emphasize informative features in support data relevant to the query data, which generalizes better on unseen subjects. For evaluation, we used the BCI competition IV 2b dataset and achieved an 9.3% accuracy improvement in the 20-shot classification task with state-of-the-art performance. Experimental results demonstrate the effectiveness of employing attention and the overall generality of our method.

Max Allan, Satoshi Kondo, Sebastian Bodenstedt et al.

In 2015 we began a sub-challenge at the EndoVis workshop at MICCAI in Munich using endoscope images of ex-vivo tissue with automatically generated annotations from robot forward kinematics and instrument CAD models. However, the limited background variation and simple motion rendered the dataset uninformative in learning about which techniques would be suitable for segmentation in real surgery. In 2017, at the same workshop in Quebec we introduced the robotic instrument segmentation dataset with 10 teams participating in the challenge to perform binary, articulating parts and type segmentation of da Vinci instruments. This challenge included realistic instrument motion and more complex porcine tissue as background and was widely addressed with modifications on U-Nets and other popular CNN architectures. In 2018 we added to the complexity by introducing a set of anatomical objects and medical devices to the segmented classes. To avoid over-complicating the challenge, we continued with porcine data which is dramatically simpler than human tissue due to the lack of fatty tissue occluding many organs.

Hugo J. Kuijf, J. Matthijs Biesbroek, Jeroen de Bresser et al.

Quantification of cerebral white matter hyperintensities (WMH) of presumed vascular origin is of key importance in many neurological research studies. Currently, measurements are often still obtained from manual segmentations on brain MR images, which is a laborious procedure. Automatic WMH segmentation methods exist, but a standardized comparison of the performance of such methods is lacking. We organized a scientific challenge, in which developers could evaluate their method on a standardized multi-center/-scanner image dataset, giving an objective comparison: the WMH Segmentation Challenge (https://wmh.isi.uu.nl/). Sixty T1+FLAIR images from three MR scanners were released with manual WMH segmentations for training. A test set of 110 images from five MR scanners was used for evaluation. Segmentation methods had to be containerized and submitted to the challenge organizers. Five evaluation metrics were used to rank the methods: (1) Dice similarity coefficient, (2) modified Hausdorff distance (95th percentile), (3) absolute log-transformed volume difference, (4) sensitivity for detecting individual lesions, and (5) F1-score for individual lesions. Additionally, methods were ranked on their inter-scanner robustness. Twenty participants submitted their method for evaluation. This paper provides a detailed analysis of the results. In brief, there is a cluster of four methods that rank significantly better than the other methods, with one clear winner. The inter-scanner robustness ranking shows that not all methods generalize to unseen scanners. The challenge remains open for future submissions and provides a public platform for method evaluation.

Deepak Ghimire, Sunghwan Jeong, Joonwhoan Lee et al.

Facial expressions are one of the most powerful, natural and immediate means for human being to communicate their emotions and intensions. Recognition of facial expression has many applications including human-computer interaction, cognitive science, human emotion analysis, personality development etc. In this paper, we propose a new method for the recognition of facial expressions from single image frame that uses combination of appearance and geometric features with support vector machines classification. In general, appearance features for the recognition of facial expressions are computed by dividing face region into regular grid (holistic representation). But, in this paper we extracted region specific appearance features by dividing the whole face region into domain specific local regions. Geometric features are also extracted from corresponding domain specific regions. In addition, important local regions are determined by using incremental search approach which results in the reduction of feature dimension and improvement in recognition accuracy. The results of facial expressions recognition using features from domain specific regions are also compared with the results obtained using holistic representation. The performance of the proposed facial expression recognition system has been validated on publicly available extended Cohn-Kanade (CK+) facial expression data sets.