Youngmin Ro

Dongheon Lee, Seokju Yun, Jaegyun Im et al.

Recent Super-Resolution~(SR) methods mainly adopt Transformers for their strong long-range modeling capability and exceptional representational capacity. However, most SR Transformers rely heavily on relative positional bias~(RPB), which prevents them from leveraging hardware-efficient attention kernels such as FlashAttention. This limitation imposes a prohibitive computational burden during both training and inference, severely restricting attempts to scale SR Transformers by enlarging the training patch size or the self-attention window. Consequently, unlike other domains that actively exploit the inherent scalability of Transformers, SR Transformers remain heavily focused on effectively utilizing limited receptive fields. In this paper, we propose Rank-factorized Implicit Neural Bias~(RIB), an alternative to RPB that enables FlashAttention in SR Transformers. Specifically, RIB approximates positional bias using low-rank implicit neural representations and concatenates them with pixel content tokens in a channel-wise manner, turning the element-wise bias addition in attention score computation into a dot-product operation. Further, we introduce a convolutional local attention and a cyclic window strategy to fully leverage the advantages of long-range interactions enabled by RIB and FlashAttention. We enlarge the window size up to \textbf{96$\times$96} while jointly scaling the training patch size and the dataset size, maximizing the benefits of Transformers in the SR task. As a result, our network achieves \textbf{35.63\,dB PSNR} on Urban100$\times$2, while reducing training and inference time by \textbf{2.1$\times$} and \textbf{2.9$\times$}, respectively, compared to the RPB-based SR Transformer~(PFT).

Seokju Yun, Youngmin Ro

We introduce Dynamic Mobile-Former(DMF), maximizes the capabilities of dynamic convolution by harmonizing it with efficient operators.Our Dynamic MobileFormer effectively utilizes the advantages of Dynamic MobileNet (MobileNet equipped with dynamic convolution) using global information from light-weight attention.A Transformer in Dynamic Mobile-Former only requires a few randomly initialized tokens to calculate global features, making it computationally efficient.And a bridge between Dynamic MobileNet and Transformer allows for bidirectional integration of local and global features.We also simplify the optimization process of vanilla dynamic convolution by splitting the convolution kernel into an input-agnostic kernel and an input-dependent kernel.This allows for optimization in a wider kernel space, resulting in enhanced capacity.By integrating lightweight attention and enhanced dynamic convolution, our Dynamic Mobile-Former achieves not only high efficiency, but also strong performance.We benchmark the Dynamic Mobile-Former on a series of vision tasks, and showcase that it achieves impressive performance on image classification, COCO detection, and instanace segmentation.For example, our DMF hits the top-1 accuracy of 79.4% on ImageNet-1K, much higher than PVT-Tiny by 4.3% with only 1/4 FLOPs.Additionally,our proposed DMF-S model performed well on challenging vision datasets such as COCO, achieving a 39.0% mAP,which is 1% higher than that of the Mobile-Former 508M model, despite using 3 GFLOPs less computations.Code and models are available at https://github.com/ysj9909/DMF

Seokju Yun, Dongheon Lee, Noori Bae et al.

As AI systems are being integrated more rapidly into diverse and complex real-world environments, the ability to perform holistic reasoning over an implicit query and an image to localize a target is becoming increasingly important. However, recent reasoning segmentation methods fail to sufficiently elicit the visual reasoning capabilities of the base mode. In this work, we present Segment Anything Reasoner (StAR), a comprehensive framework that refines the design space from multiple perspectives-including parameter-tuning scheme, reward functions, learning strategies and answer format-and achieves substantial improvements over recent baselines. In addition, for the first time, we successfully introduce parallel test-time scaling to the segmentation task, pushing the performance boundary even further. To extend the scope and depth of reasoning covered by existing benchmark, we also construct the ReasonSeg-X, which compactly defines reasoning types and includes samples that require deeper reasoning. Leveraging this dataset, we train StAR with a rollout-expanded selective-tuning approach to activate the base model's latent reasoning capabilities, and establish a rigorous benchmark for systematic, fine-grained evaluation of advanced methods. With only 5k training samples, StAR achieves significant gains over its base counterparts across extensive benchmarks, demonstrating that our method effectively brings dormant reasoning competence to the surface.

Chanseul Cho, Seokju Yun, Jeaseong Jeon et al.

Domain Generalized Semantic Segmentation (DGSS) aims to maintain robust performance across unseen target domains. Vision Foundation Models (VFMs) offer rich multi-domain knowledge that can enhance generalization. However, strategies for actively exploiting the rich subspace structures within VFMs remain under-explored, with many existing methods focusing primarily on preserving pre-trained knowledge. Furthermore, their LoRA components often suffer from limited representational diversity and inefficient parameter utilization. We propose RecycleLoRA, which addresses both challenges by employing Rank-Revealing QR Decomposition (RRQR) to systematically exploit VFM's subspace structures and enhance LoRA's representational richness. Our main adapter leverages minor subspace directions identified by RRQR to learn diverse and independent features, achieving competitive performance even when used alone. We further introduce a sub adapter that carefully refines major directions with minimal adjustments, providing complementary improvements to the main adapter's strong baseline performance. This design enables the dual adapters to learn distinct representations without requiring additional regularization losses. Our systematic exploitation of pre-trained subspace structures through RRQR-based initialization leads to superior domain generalization performance. RecycleLoRA achieves state-of-the-art performance on both synthetic-to-real generalization and real-to-real generalization tasks without complex architectures or additional inference latency.

Youngmin Ro, Jin Young Choi

Existing fine-tuning methods use a single learning rate over all layers. In this paper, first, we discuss that trends of layer-wise weight variations by fine-tuning using a single learning rate do not match the well-known notion that lower-level layers extract general features and higher-level layers extract specific features. Based on our discussion, we propose an algorithm that improves fine-tuning performance and reduces network complexity through layer-wise pruning and auto-tuning of layer-wise learning rates. The proposed algorithm has verified the effectiveness by achieving state-of-the-art performance on the image retrieval benchmark datasets (CUB-200, Cars-196, Stanford online product, and Inshop). Code is available at https://github.com/youngminPIL/AutoLR.

Dongheon Lee, Seokju Yun, Youngmin Ro

For Image Super-Resolution (SR), it is common to train and evaluate scale-specific models composed of an encoder and upsampler for each targeted scale. Consequently, many SR studies encounter substantial training times and complex deployment requirements. In this paper, we address this limitation by training and evaluating multiple scales simultaneously. Notably, we observe that encoder features are similar across scales and that the Sub-Pixel Convolution (SPConv), widely-used scale-specific upsampler, exhibits strong inter-scale correlations in its functionality. Building on these insights, we propose a multi-scale framework that employs a single encoder in conjunction with Implicit Grid Convolution (IGConv), our novel upsampler, which unifies SPConv across all scales within a single module. Extensive experiments demonstrate that our framework achieves comparable performance to existing fixed-scale methods while reducing the training budget and stored parameters three-fold and maintaining the same latency. Additionally, we propose IGConv$^{+}$ to improve performance further by addressing spectral bias and allowing input-dependent upsampling and ensembled prediction. As a result, ATD-IGConv$^{+}$ achieves a notable 0.21dB improvement in PSNR on Urban100$\times$4, while also reducing the training budget, stored parameters, and inference cost compared to the existing ATD.

Seokju Yun, Youngmin Ro

Recently, efficient Vision Transformers have shown great performance with low latency on resource-constrained devices. Conventionally, they use 4x4 patch embeddings and a 4-stage structure at the macro level, while utilizing sophisticated attention with multi-head configuration at the micro level. This paper aims to address computational redundancy at all design levels in a memory-efficient manner. We discover that using larger-stride patchify stem not only reduces memory access costs but also achieves competitive performance by leveraging token representations with reduced spatial redundancy from the early stages. Furthermore, our preliminary analyses suggest that attention layers in the early stages can be substituted with convolutions, and several attention heads in the latter stages are computationally redundant. To handle this, we introduce a single-head attention module that inherently prevents head redundancy and simultaneously boosts accuracy by parallelly combining global and local information. Building upon our solutions, we introduce SHViT, a Single-Head Vision Transformer that obtains the state-of-the-art speed-accuracy tradeoff. For example, on ImageNet-1k, our SHViT-S4 is 3.3x, 8.1x, and 2.4x faster than MobileViTv2 x1.0 on GPU, CPU, and iPhone12 mobile device, respectively, while being 1.3% more accurate. For object detection and instance segmentation on MS COCO using Mask-RCNN head, our model achieves performance comparable to FastViT-SA12 while exhibiting 3.8x and 2.0x lower backbone latency on GPU and mobile device, respectively.

Seokju Yun, Seunghye Chae, Dongheon Lee et al.

Domain generalization (DG) aims to adapt a model using one or multiple source domains to ensure robust performance in unseen target domains. Recently, Parameter-Efficient Fine-Tuning (PEFT) of foundation models has shown promising results in the context of DG problem. Nevertheless, existing PEFT methods still struggle to strike a balance between preserving generalizable components of the pre-trained model and learning task-specific features. To gain insights into the distribution of generalizable components, we begin by analyzing the pre-trained weights through the lens of singular value decomposition. Building on these insights, we introduce Singular Value Decomposed Minor Components Adaptation (SoMA), an approach that selectively tunes minor singular components while keeping the residual parts frozen. SoMA effectively retains the generalization ability of the pre-trained model while efficiently acquiring task-specific skills. Moreover, we freeze domain-generalizable blocks and employ an annealing weight decay strategy, thereby achieving an optimal balance in the delicate trade-off between generalizability and discriminability. SoMA attains state-of-the-art results on multiple benchmarks that span both domain generalized semantic segmentation to domain generalized object detection. In addition, our methods introduce no additional inference overhead or regularization loss, maintain compatibility with any backbone or head, and are designed to be versatile, allowing easy integration into a wide range of tasks.

Dongheon Lee, Seokju Yun, Youngmin Ro

Recently, in the super-resolution (SR) domain, transformers have outperformed CNNs with fewer FLOPs and fewer parameters since they can deal with long-range dependency and adaptively adjust weights based on instance. In this paper, we demonstrate that CNNs, although less focused on in the current SR domain, surpass Transformers in direct efficiency measures. By incorporating the advantages of Transformers into CNNs, we aim to achieve both computational efficiency and enhanced performance. However, using a large kernel in the SR domain, which mainly processes large images, incurs a large computational overhead. To overcome this, we propose novel approaches to employing the large kernel, which can reduce latency by 86\% compared to the naive large kernel, and leverage an Element-wise Attention module to imitate instance-dependent weights. As a result, we introduce Partial Large Kernel CNNs for Efficient Super-Resolution (PLKSR), which achieves state-of-the-art performance on four datasets at a scale of $\times$4, with reductions of 68.1\% in latency and 80.2\% in maximum GPU memory occupancy compared to SRFormer-light.

Seungjae Moon, Seunghyun Oh, Youngmin Ro

Training-free open-vocabulary semantic segmentation(TF-OVSS) has recently attracted attention for its ability to perform dense prediction by leveraging the pretrained knowledge of large vision and vision-language models, without requiring additional training. However, due to the limited input resolution of these pretrained encoders, existing TF-OVSS methods commonly adopt a sliding-window strategy that processes cropped sub-images independently. While effective for managing high-resolution inputs, this approach prevents global attention over the full image, leading to fragmented feature representations and limited contextual reasoning. We propose OV-Stitcher, a training-free framework that addresses this limitation by stitching fragmented sub-image features directly within the final encoder block. By reconstructing attention representations from fragmented sub-image features, OV-Stitcher enables global attention within the final encoder block, producing coherent context aggregation and spatially consistent, semantically aligned segmentation maps. Extensive evaluations across eight benchmarks demonstrate that OV-Stitcher establishes a scalable and effective solution for open-vocabulary segmentation, achieving a notable improvement in mean Intersection over Union(mIoU) from 48.7 to 50.7 compared with prior training-free baselines.

Dongheon Lee, Seokju Yun, Youngmin Ro

In this paper, we tackle the high computational overhead of Transformers for efficient image super-resolution~(SR). Motivated by the observations of self-attention's inter-layer repetition, we introduce a convolutionized self-attention module named Convolutional Attention~(ConvAttn) that emulates self-attention's long-range modeling capability and instance-dependent weighting with a single shared large kernel and dynamic kernels. By utilizing the ConvAttn module, we significantly reduce the reliance on self-attention and its involved memory-bound operations while maintaining the representational capability of Transformers. Furthermore, we overcome the challenge of integrating flash attention into the lightweight SR regime, effectively mitigating self-attention's inherent memory bottleneck. We scale up the window size to 32$\times$32 with flash attention rather than proposing an intricate self-attention module, significantly improving PSNR by 0.31dB on Urban100$\times$2 while reducing latency and memory usage by 16$\times$ and 12.2$\times$. Building on these approaches, our proposed network, termed Emulating Self-attention with Convolution~(ESC), notably improves PSNR by 0.27 dB on Urban100$\times$4 compared to HiT-SRF, reducing the latency and memory usage by 3.7$\times$ and 6.2$\times$, respectively. Extensive experiments demonstrate that our ESC maintains the ability for long-range modeling, data scalability, and the representational power of Transformers despite most self-attention being replaced by the ConvAttn module.

Seokju Yun, Dongheon Lee, Youngmin Ro

Transformer, composed of self-attention and Feed-Forward Network, has revolutionized the landscape of network design across various vision tasks. While self-attention is extensively explored as a key factor in performance, FFN has received little attention. FFN is a versatile operator seamlessly integrated into nearly all AI models to effectively harness rich representations. Recent works also show that FFN functions like key-value memories. Thus, akin to the query-key-value mechanism within self-attention, FFN can be viewed as a memory network, where the input serves as query and the two projection weights operate as keys and values, respectively. Based on these observations, we hypothesize that the importance lies in query-key-value framework itself for competitive performance. To verify this, we propose converting self-attention into a more FFN-like efficient token mixer with only convolutions while retaining query-key-value framework, namely FFNification. Specifically, FFNification replaces query-key-value interactions with large kernel convolutions and adopts GELU activation function instead of softmax. The derived token mixer, FFNified attention, serves as key-value memories for detecting locally distributed spatial patterns, and operates in the opposite dimension to the ConvNeXt block within each corresponding sub-operation of the query-key-value framework. Building upon the above two modules, we present a family of Fast-Forward Networks (FFNet). Despite being composed of only simple operators, FFNet outperforms sophisticated and highly specialized methods in each domain, with notable efficiency gains. These results validate our hypothesis, leading us to propose MetaMixer, a general mixer architecture that does not specify sub-operations within the query-key-value framework.

Dongheon Lee, Seungmyong Jeong, Youngmin Ro

Numerical models have long been used to understand geoscientific phenomena, including tidal currents, crucial for renewable energy production and coastal engineering. However, their computational cost hinders generating data of varying resolutions. As an alternative, deep learning-based downscaling methods have gained traction due to their faster inference speeds. But most of them are limited to only inference fixed scale and overlook important characteristics of target geoscientific data. In this paper, we propose a novel downscaling framework for tidal current data, addressing its unique characteristics, which are dissimilar to images: heterogeneity and local dependency. Moreover, our framework can generate any arbitrary-scale output utilizing a continuous representation model. Our proposed framework demonstrates significantly improved flow velocity predictions by 93.21% (MSE) and 63.85% (MAE) compared to the Baseline model while achieving a remarkable 33.2% reduction in FLOPs.

Yonghyun Jeong, Doyeon Kim, Youngmin Ro et al.

Various deepfake detectors have been proposed, but challenges still exist to detect images of unknown categories or GAN models outside of the training settings. Such issues arise from the overfitting issue, which we discover from our own analysis and the previous studies to originate from the frequency-level artifacts in generated images. We find that ignoring the frequency-level artifacts can improve the detector's generalization across various GAN models, but it can reduce the model's performance for the trained GAN models. Thus, we design a framework to generalize the deepfake detector for both the known and unseen GAN models. Our framework generates the frequency-level perturbation maps to make the generated images indistinguishable from the real images. By updating the deepfake detector along with the training of the perturbation generator, our model is trained to detect the frequency-level artifacts at the initial iterations and consider the image-level irregularities at the last iterations. For experiments, we design new test scenarios varying from the training settings in GAN models, color manipulations, and object categories. Numerous experiments validate the state-of-the-art performance of our deepfake detector.

Yonghyun Jeong, Doyeon Kim, Pyounggeon Kim et al.

Although the recent advancement in generative models brings diverse advantages to society, it can also be abused with malicious purposes, such as fraud, defamation, and fake news. To prevent such cases, vigorous research is conducted to distinguish the generated images from the real images, but challenges still remain to distinguish the unseen generated images outside of the training settings. Such limitations occur due to data dependency arising from the model's overfitting issue to the training data generated by specific GANs. To overcome this issue, we adopt a self-supervised scheme to propose a novel framework. Our proposed method is composed of the artificial fingerprint generator reconstructing the high-quality artificial fingerprints of GAN images for detailed analysis, and the GAN detector distinguishing GAN images by learning the reconstructed artificial fingerprints. To improve the generalization of the artificial fingerprint generator, we build multiple autoencoders with different numbers of upconvolution layers. With numerous ablation studies, the robust generalization of our method is validated by outperforming the generalization of the previous state-of-the-art algorithms, even without utilizing the GAN images of the training dataset.

Yonghyun Jeong, Jooyoung Choi, Sungwon Kim et al.

In this work, we present Facial Identity Controllable GAN (FICGAN) for not only generating high-quality de-identified face images with ensured privacy protection, but also detailed controllability on attribute preservation for enhanced data utility. We tackle the less-explored yet desired functionality in face de-identification based on the two factors. First, we focus on the challenging issue to obtain a high level of privacy protection in the de-identification task while uncompromising the image quality. Second, we analyze the facial attributes related to identity and non-identity and explore the trade-off between the degree of face de-identification and preservation of the source attributes for enhanced data utility. Based on the analysis, we develop Facial Identity Controllable GAN (FICGAN), an autoencoder-based conditional generative model that learns to disentangle the identity attributes from non-identity attributes on a face image. By applying the manifold k-same algorithm to satisfy k-anonymity for strengthened security, our method achieves enhanced privacy protection in de-identified face images. Numerous experiments demonstrate that our model outperforms others in various scenarios of face de-identification.

Youngmin Ro, Jongwon Choi, Dae Ung Jo et al.

In person re-identification (ReID) task, because of its shortage of trainable dataset, it is common to utilize fine-tuning method using a classification network pre-trained on a large dataset. However, it is relatively difficult to sufficiently fine-tune the low-level layers of the network due to the gradient vanishing problem. In this work, we propose a novel fine-tuning strategy that allows low-level layers to be sufficiently trained by rolling back the weights of high-level layers to their initial pre-trained weights. Our strategy alleviates the problem of gradient vanishing in low-level layers and robustly trains the low-level layers to fit the ReID dataset, thereby increasing the performance of ReID tasks. The improved performance of the proposed strategy is validated via several experiments. Furthermore, without any add-ons such as pose estimation or segmentation, our strategy exhibits state-of-the-art performance using only vanilla deep convolutional neural network architecture.