PyTorchKR Papers

최신 논문

총 3,696편
태그·기간 등 상세 필터

카테고리 탐색

로그인 후 이 태그 구독

arXiv:1805.11529

On Low-Resolution Face Recognition in the Wild: Comparisons and New Techniques

Although face recognition systems have achieved impressive performance in recent years, the low-resolution face recognition (LRFR) task remains challenging, especially when the LR faces are captured under non-ideal conditions, as is common in surveillance-based applications. Faces captured in such conditions are often contaminated by blur, nonuniform lighting, and nonfrontal face pose. In this paper, we analyze face recognition techniques using data captured under low-quality conditions in the wild. We provide a comprehensive analysis of experimental results for two of the most important applications in real surveillance applications, and demonstrate practical approaches to handle both cases that show promising performance. The following three contributions are made: {\em (i)} we conduct experiments to evaluate super-resolution methods for low-resolution face recognition; {\em (ii)} we study face re-identification on various public face datasets including real surveillance and low-resolution subsets of large-scale datasets, present a baseline result for several deep learning based approaches, and improve them by introducing a GAN pre-training approach and fully convolutional architecture; and {\em (iii)} we explore low-resolution face identification by employing a state-of-the-art supervised discriminative learning approach. Evaluations are conducted on challenging portions of the SCFace and UCCSface datasets.

arXiv:1805.11519

Face Recognition in Low Quality Images: A Survey

Low-resolution face recognition (LRFR) has received increasing attention over the past few years. Its applications lie widely in the real-world environment when high-resolution or high-quality images are hard to capture. One of the biggest demands for LRFR technologies is video surveillance. As the the number of surveillance cameras in the city increases, the videos that captured will need to be processed automatically. However, those videos or images are usually captured with large standoffs, arbitrary illumination condition, and diverse angles of view. Faces in these images are generally small in size. Several studies addressed this problem employed techniques like super resolution, deblurring, or learning a relationship between different resolution domains. In this paper, we provide a comprehensive review of approaches to low-resolution face recognition in the past five years. First, a general problem definition is given. Later, systematically analysis of the works on this topic is presented by catogory. In addition to describing the methods, we also focus on datasets and experiment settings. We further address the related works on unconstrained low-resolution face recognition and compare them with the result that use synthetic low-resolution data. Finally, we summarized the general limitations and speculate a priorities for the future effort.

arXiv:1802.09987공식 코드 · Papers with Code

Multi-View Silhouette and Depth Decomposition for High Resolution 3D Object Representation

We consider the problem of scaling deep generative shape models to high-resolution. Drawing motivation from the canonical view representation of objects, we introduce a novel method for the fast up-sampling of 3D objects in voxel space through networks that perform super-resolution on the six orthographic depth projections. This allows us to generate high-resolution objects with more efficient scaling than methods which work directly in 3D. We decompose the problem of 2D depth super-resolution into silhouette and depth prediction to capture both structure and fine detail. This allows our method to generate sharp edges more easily than an individual network. We evaluate our work on multiple experiments concerning high-resolution 3D objects, and show our system is capable of accurately predicting novel objects at resolutions as large as 512$\mathbf{\times}$512$\mathbf{\times}$512 -- the highest resolution reported for this task. We achieve state-of-the-art performance on 3D object reconstruction from RGB images on the ShapeNet dataset, and further demonstrate the first effective 3D super-resolution method.

arXiv:1806.00874

Patch-Based Image Hallucination for Super Resolution with Detail Reconstruction from Similar Sample Images

Image hallucination and super-resolution have been studied for decades, and many approaches have been proposed to upsample low-resolution images using information from the images themselves, multiple example images, or large image databases. However, most of this work has focused exclusively on small magnification levels because the algorithms simply sharpen the blurry edges in the upsampled images - no actual new detail is typically reconstructed in the final result. In this paper, we present a patch-based algorithm for image hallucination which, for the first time, properly synthesizes novel high frequency detail. To do this, we pose the synthesis problem as a patch-based optimization which inserts coherent, high-frequency detail from contextually-similar images of the same physical scene/subject provided from either a personal image collection or a large online database. The resulting image is visually plausible and contains coherent high frequency information. We demonstrate the robustness of our algorithm by testing it on a large number of images and show that its performance is considerably superior to all state-of-the-art approaches, a result that is verified to be statistically significant through a randomized user study.

arXiv:1610.09600

Super-resolution estimation of cyclic arrival rates

Exploiting the fact that most arrival processes exhibit cyclic behaviour, we propose a simple procedure for estimating the intensity of a nonhomogeneous Poisson process. The estimator is the super-resolution analogue to Shao 2010 and Shao & Lii 2011, which is a sum of $p$ sinusoids where $p$ and the frequency, amplitude, and phase of each wave are not known and need to be estimated. This results in an interpretable yet flexible specification that is suitable for use in modelling as well as in high resolution simulations. Our estimation procedure sits in between classic periodogram methods and atomic/total variation norm thresholding. Through a novel use of window functions in the point process domain, our approach attains super-resolution without semidefinite programming. Under suitable conditions, finite sample guarantees can be derived for our procedure. These resolve some open questions and expand existing results in spectral estimation literature.

arXiv:1806.01576

Adaptive Importance Learning for Improving Lightweight Image Super-resolution Network

Deep neural networks have achieved remarkable success in single image super-resolution (SISR). The computing and memory requirements of these methods have hindered their application to broad classes of real devices with limited computing power, however. One approach to this problem has been lightweight network architectures that bal- ance the super-resolution performance and the computation burden. In this study, we revisit this problem from an orthog- onal view, and propose a novel learning strategy to maxi- mize the pixel-wise fitting capacity of a given lightweight network architecture. Considering that the initial capacity of the lightweight network is very limited, we present an adaptive importance learning scheme for SISR that trains the network with an easy-to-complex paradigm by dynam- ically updating the importance of image pixels on the basis of the training loss. Specifically, we formulate the network training and the importance learning into a joint optimization problem. With a carefully designed importance penalty function, the importance of individual pixels can be gradu- ally increased through solving a convex optimization problem. The training process thus begins with pixels that are easy to reconstruct, and gradually proceeds to more complex pixels as fitting improves.

arXiv:1806.02658

Super-Resolution using Convolutional Neural Networks without Any Checkerboard Artifacts

It is well-known that a number of excellent super-resolution (SR) methods using convolutional neural networks (CNNs) generate checkerboard artifacts. A condition to avoid the checkerboard artifacts is proposed in this paper. So far, checkerboard artifacts have been mainly studied for linear multirate systems, but the condition to avoid checkerboard artifacts can not be applied to CNNs due to the non-linearity of CNNs. We extend the avoiding condition for CNNs, and apply the proposed structure to some typical SR methods to confirm the effectiveness of the new scheme. Experiment results demonstrate that the proposed structure can perfectly avoid to generate checkerboard artifacts under two loss conditions: mean square error and perceptual loss, while keeping excellent properties that the SR methods have.

arXiv:1806.02919공식 코드 · Papers with Code

Non-Local Recurrent Network for Image Restoration

Many classic methods have shown non-local self-similarity in natural images to be an effective prior for image restoration. However, it remains unclear and challenging to make use of this intrinsic property via deep networks. In this paper, we propose a non-local recurrent network (NLRN) as the first attempt to incorporate non-local operations into a recurrent neural network (RNN) for image restoration. The main contributions of this work are: (1) Unlike existing methods that measure self-similarity in an isolated manner, the proposed non-local module can be flexibly integrated into existing deep networks for end-to-end training to capture deep feature correlation between each location and its neighborhood. (2) We fully employ the RNN structure for its parameter efficiency and allow deep feature correlation to be propagated along adjacent recurrent states. This new design boosts robustness against inaccurate correlation estimation due to severely degraded images. (3) We show that it is essential to maintain a confined neighborhood for computing deep feature correlation given degraded images. This is in contrast to existing practice that deploys the whole image. Extensive experiments on both image denoising and super-resolution tasks are conducted. Thanks to the recurrent non-local operations and correlation propagation, the proposed NLRN achieves superior results to state-of-the-art methods with much fewer parameters.

arXiv:1803.01417

Efficient and Accurate MRI Super-Resolution using a Generative Adversarial Network and 3D Multi-Level Densely Connected Network

High-resolution (HR) magnetic resonance images (MRI) provide detailed anatomical information important for clinical application and quantitative image analysis. However, HR MRI conventionally comes at the cost of longer scan time, smaller spatial coverage, and lower signal-to-noise ratio (SNR). Recent studies have shown that single image super-resolution (SISR), a technique to recover HR details from one single low-resolution (LR) input image, could provide high-quality image details with the help of advanced deep convolutional neural networks (CNN). However, deep neural networks consume memory heavily and run slowly, especially in 3D settings. In this paper, we propose a novel 3D neural network design, namely a multi-level densely connected super-resolution network (mDCSRN) with generative adversarial network (GAN)-guided training. The mDCSRN quickly trains and inferences and the GAN promotes realistic output hardly distinguishable from original HR images. Our results from experiments on a dataset with 1,113 subjects show that our new architecture beats other popular deep learning methods in recovering 4x resolution-downgraded im-ages and runs 6x faster.

arXiv:1709.09422

Light field super resolution through controlled micro-shifts of light field sensor

Light field cameras enable new capabilities, such as post-capture refocusing and aperture control, through capturing directional and spatial distribution of light rays in space. Micro-lens array based light field camera design is often preferred due to its light transmission efficiency, cost-effectiveness and compactness. One drawback of the micro-lens array based light field cameras is low spatial resolution due to the fact that a single sensor is shared to capture both spatial and angular information. To address the low spatial resolution issue, we present a light field imaging approach, where multiple light fields are captured and fused to improve the spatial resolution. For each capture, the light field sensor is shifted by a pre-determined fraction of a micro-lens size using an XY translation stage for optimal performance.

arXiv:1705.09992공식 코드 · Papers with Code

LAP: a Linearize and Project Method for Solving Inverse Problems with Coupled Variables

Many inverse problems involve two or more sets of variables that represent different physical quantities but are tightly coupled with each other. For example, image super-resolution requires joint estimation of the image and motion parameters from noisy measurements. Exploiting this structure is key for efficiently solving these large-scale optimization problems, which are often ill-conditioned. In this paper, we present a new method called Linearize And Project (LAP) that offers a flexible framework for solving inverse problems with coupled variables. LAP is most promising for cases when the subproblem corresponding to one of the variables is considerably easier to solve than the other. LAP is based on a Gauss-Newton method, and thus after linearizing the residual, it eliminates one block of variables through projection. Due to the linearization, this block can be chosen freely. Further, LAP supports direct, iterative, and hybrid regularization as well as constraints. Therefore LAP is attractive, e.g., for ill-posed imaging problems. These traits differentiate LAP from common alternatives for this type of problem such as variable projection (VarPro) and block coordinate descent (BCD). Our numerical experiments compare the performance of LAP to BCD and VarPro using three coupled problems whose forward operators are linear with respect to one block and nonlinear for the other set of variables.

arXiv:1806.05764

Generative Adversarial Networks and Perceptual Losses for Video Super-Resolution

Video super-resolution (VSR) has become one of the most critical problems in video processing. In the deep learning literature, recent works have shown the benefits of using adversarial-based and perceptual losses to improve the performance on various image restoration tasks; however, these have yet to be applied for video super-resolution. In this work, we propose a Generative Adversarial Network(GAN)-based formulation for VSR. We introduce a new generator network optimized for the VSR problem, named VSRResNet, along with a new discriminator architecture to properly guide VSRResNet during the GAN training. We further enhance our VSR GAN formulation with two regularizers, a distance loss in feature-space and pixel-space, to obtain our final VSRResFeatGAN model. We show that pre-training our generator with the Mean-Squared-Error loss only quantitatively surpasses the current state-of-the-art VSR models. Finally, we employ the PercepDist metric (Zhang et al., 2018) to compare state-of-the-art VSR models. We show that this metric more accurately evaluates the perceptual quality of SR solutions obtained from neural networks, compared with the commonly used PSNR/SSIM metrics. Finally, we show that our proposed model, the VSRResFeatGAN model, outperforms current state-of-the-art SR models, both quantitatively and qualitatively.

arXiv:1806.03575

Accurate Spectral Super-resolution from Single RGB Image Using Multi-scale CNN

Different from traditional hyperspectral super-resolution approaches that focus on improving the spatial resolution, spectral super-resolution aims at producing a high-resolution hyperspectral image from the RGB observation with super-resolution in spectral domain. However, it is challenging to accurately reconstruct a high-dimensional continuous spectrum from three discrete intensity values at each pixel, since too much information is lost during the procedure where the latent hyperspectral image is downsampled (e.g., with x10 scaling factor) in spectral domain to produce an RGB observation. To address this problem, we present a multi-scale deep convolutional neural network (CNN) to explicitly map the input RGB image into a hyperspectral image. Through symmetrically downsampling and upsampling the intermediate feature maps in a cascading paradigm, the local and non-local image information can be jointly encoded for spectral representation, ultimately improving the spectral reconstruction accuracy. Extensive experiments on a large hyperspectral dataset demonstrate the effectiveness of the proposed method.

arXiv:1805.03383공식 코드 · Papers with Code

New Techniques for Preserving Global Structure and Denoising with Low Information Loss in Single-Image Super-Resolution

This work identifies and addresses two important technical challenges in single-image super-resolution: (1) how to upsample an image without magnifying noise and (2) how to preserve large scale structure when upsampling. We summarize the techniques we developed for our second place entry in Track 1 (Bicubic Downsampling), seventh place entry in Track 2 (Realistic Adverse Conditions), and seventh place entry in Track 3 (Realistic difficult) in the 2018 NTIRE Super-Resolution Challenge. Furthermore, we present new neural network architectures that specifically address the two challenges listed above: denoising and preservation of large-scale structure.

arXiv:1802.05751

Image Transformer

Image generation has been successfully cast as an autoregressive sequence generation or transformation problem. Recent work has shown that self-attention is an effective way of modeling textual sequences. In this work, we generalize a recently proposed model architecture based on self-attention, the Transformer, to a sequence modeling formulation of image generation with a tractable likelihood. By restricting the self-attention mechanism to attend to local neighborhoods we significantly increase the size of images the model can process in practice, despite maintaining significantly larger receptive fields per layer than typical convolutional neural networks. While conceptually simple, our generative models significantly outperform the current state of the art in image generation on ImageNet, improving the best published negative log-likelihood on ImageNet from 3.83 to 3.77. We also present results on image super-resolution with a large magnification ratio, applying an encoder-decoder configuration of our architecture. In a human evaluation study, we find that images generated by our super-resolution model fool human observers three times more often than the previous state of the art.

arXiv:1701.05652

Dual Recovery Network with Online Compensation for Image Super-Resolution

Image super-resolution (SR) methods essentially lead to a loss of some high-frequency (HF) information when predicting high-resolution (HR) images from low-resolution (LR) images without using external references. To address this issue, we additionally utilize online retrieved data to facilitate image SR in a unified deep framework. A novel dual high-frequency recovery network (DHN) is proposed to predict an HR image with three parts: an LR image, an internal inferred HF (IHF) map (HF missing part inferred solely from the LR image) and an external extracted HF (EHF) map. In particular, we infer the HF information based on both the LR image and similar HR references which are retrieved online. For the EHF map, we align the references with affine transformation and then in the aligned references, part of HF signals are extracted by the proposed DHN to compensate for the HF loss. Extensive experimental results demonstrate that our DHN achieves notably better performance than state-of-the-art SR methods.

arXiv:2610.15752v1cs.CV

동결된 히스토리: 학습 기반 실시간 슈퍼샘플링에서 시간적 손실이 누적을 중단시키는 시점

Frozen History: When Temporal Losses Stop Accumulation in Learned Real-Time Supersampling

학습(training) 기반 실시간 슈퍼샘플링(supersampling)은 모션 벡터(motion vector)를 따라 지터링된(jittered) 저해상도 샘플들을 누적합니다. 많은 방법들이 깜빡임(flicker)을 줄이기 위해 시간적 일관성 손실(temporal consistency loss)을 사용합니다. 본 연구는 정확한 정합(registration)이 주어지면 이 손실이 누적을 억제할 수 있음을 보여줍니다. 정적 콘텐츠에서 네트워크는 거의 시간적 비용 없이 워핑된 히스토리를 그대로 복사합니다. 모션 벡터의 카메라 지터는 이 비용을 높여 해당 효과를 감출 수 있습니다. 본 논문에서는 학습 과정에서 은폐될 수 있는 네 가지 정합 오차로 지터를 포함하는 모션 벡터, 두 해상도 격자 모두에 사용되는 단일 모션 필드, 지터링된 깊이(depth), 정합되지 않은 스킵 경로(skip path)를 식별합니다. 모델(model) 프리 오라클을 통해 정확한 정합이 정적 픽셀에서 동결(freezing) 비용을 2차수(two orders of magnitude) 더 낮춘다는 것을 입증합니다. 제어된 학습 실험에서는 정합이 PSNR에 미치는 시간적 가중치(weight)의 영향을 반전시키는 것으로 나타났습니다. 정확한 정합 하에서 가중치를 늘리면 정확도와 인지 품질을 희생하면서 픽셀 분산을 줄이며, 가중치 0.25에서 정적 프레임의 손실이 0.7~0.9 dB에 달합니다. 이러한 반전 현상은 세 가지 네트워크 제품군과 두 가지 스케일 전반에서 반복되며, 해당 손실은 지터가 없는 STSS 데이터에서도 정확도를 저하시킵니다. 작은 시간적 가중치와 정규화된 샘플 누적기를 결합한 단순 정합 네트워크는 1080p로의 $\times$2, $\times$3, $\times$4 업스케일링에서 기존에 발표된 최고 QRISP PSNR 대비 0.34~0.62 dB 향상된 성능을 보여줍니다. $\times$2에서 정적 픽셀 표준편차는 DLSS 2의 1/4 수준입니다. 재학습 없이도 Arm의 지터링된 Bistro 캡처에서 고전적 시간적 업샘플링 대비 최소 4.4 dB 뛰어난 성능을 보입니다. 가장 정확한 $\times$4 모델은 H200에서 1080p 프레임당 2.03 ms로 구동됩니다. 코드: https://github.com/CV-AC/Frozen_history

arXiv:2610.04781v2cs.CV

적절한 잠재 공간에서의 초해상도: 동결된 비전 파운데이션 기저

Super-Resolution in The Right Latent Space: A Frozen Vision-Foundation Substrate

이미지 잠재 공간(latent space)에서 고해상도의 자연스럽고 선명한 이미지 임베딩은 하나의 다양체(manifold)를 형성합니다. 고해상도 이미지가 열화되면 해당 임베딩은 이 다양체 밖으로 밀려납니다. 따라서 실세계 초해상도(SR, super-resolution)는 열화된 임베딩을 다시 이 다양체로 매핑하는 작업이 됩니다. 이때 다양체 위의 임의의 위치가 아니라 입력이 여전히 담고 있는 시맨틱과 픽셀 세부 정보를 모두 보존하는 지점으로 매핑해야 합니다. 기존에 발표된 모든 방법은 이러한 매핑을 재구성 지향적 잠재 공간이나 픽셀 공간에서 구현합니다. 저자들은 이러한 공간이 SR에 적합하지 않은 기저(substrate)라고 주장합니다. 저해상도 및 열화된 이미지는 다양체에서 멀리 떨어진 곳에 임베딩되므로 매핑이 어렵고 연산 비용이 많이 듭니다. 또한 이러한 기저에는 시맨틱 정보가 부족하여 다양체 상의 정확한 지점을 찾아가기 어려우며, 열화가 심할 때 심각한 환각(hallucination)을 유발합니다. 따라서 적절한 잠재 공간에서 복원하는 것이 SR 작업에서 매우 중요합니다. 본 논문에서는 동결된 DINOv3-L의 23개 레이어를 융합한 잠재 공간이 SR 작업을 더 쉽게 만드는 공간 중 하나임을 보입니다. 열화된 이미지가 다양체 근처에 임베딩되며, 나아가 이 기저는 픽셀 기록부터 열화에 강건한 시맨틱에 이르기까지 계층적 정보를 포함하고 있어 모델이 다양체 상의 충실한 지점을 찾도록 유도합니다. 이 기저 위에서 415M 디코더는 재구성 및 적대적 목표 함수 하에 학습되어 열화된 임베딩을 다시 매핑하고 단일 패스로 픽셀 공간으로 디코딩합니다. 그 결과 제안하는 모델인 RAESR은 단일 H20 GPU에서 512x512 이미지당 37ms의 속도로 RealSR, DRealSR, LSDIR, DIV2K-Val 벤치마크에서 최신 적대적 및 확산 기반 복원기들 중 가장 우수한 충실도-지각 트레이드오프(fidelity-perception trade-off)를 달성합니다. 동일한 학습 방식으로 기저를 VAE 잠재 공간으로 교체했을 때는 모든 지표에서 성능이 저하되었습니다.

arXiv:2412.03758v4cs.CV

자기회귀 모델에는 구조적 레지스터가 필요하다: 의미론적 예측을 통한 일관된 주행 비디오 연속 생성 향상

Auto-Regressive Models Need Structural Registers: Semantic Foresight Improves Coherent Driving Video Continuation

전방 주행 비디오 연속 생성(video continuation)은 정교한 월드 모델을 구축하기 위한 핵심 구성 요소입니다. 그러나 장기적인 일관성을 유지하는 과정에서 자기회귀(auto-regressive) 프로세스 중 '생성적 퇴화(generative degeneration)'를 완화해야 하는 근본적인 난제에 직면하게 됩니다. 이 현상은 시퀀스가 길어짐에 따라 월드 모델의 제한된 어텐션(attention) 용량이 엔트로피가 높은 픽셀 세부 정보로 포화되어, 전반적인 장면 구조와 움직임 강도가 붕괴하면서 발생합니다. 이를 해결하기 위해 본 논문에서는 강제된 의미론적 예측(semantic foresight) 메커니즘을 갖춘 대규모 비전 모델 아키텍처인 STRIDE(STructural RegIsters for Decoupled Extrapolation)를 제안합니다. STRIDE는 시맨틱 토큰과 RGB 토큰의 교차 생성을 통해 비디오 연속 생성을 두 가지 하위 작업, 즉 낮은 엔트로피의 시맨틱 맵 예측과 해당 구조를 조건으로 하는 고주파 시각적 외형 생성으로 분리합니다. 시맨틱 토큰은 장면의 장기적 맥락과 월드 다이내믹스를 견고하게 유지하는 전용 메모리 슬롯인 '구조적 레지스터(structural registers)' 역할을 명시적으로 수행합니다. 나아가 아키텍처적으로 분리된 플로우 매칭(flow matching) 기반 초해상화 모듈을 사용하여 최종 시각적 충실도를 향상시킵니다. 자율주행 시나리오에 대한 광범위한 실험을 통해 STRIDE가 생성적 퇴화를 효과적으로 완화하며, 강력한 일반화 성능을 바탕으로 분(minute) 단위의 시간적으로 일관된 주행 비디오 연속 생성을 달성함을 입증합니다.

arXiv:2606.15158v3cs.CV

RefGC-SR$^2$: 참조 기반 생성형 AI 콘텐츠 초해상화 및 정제

RefGC-SR$^2$: Reference-guided Super-Resolution and Refinement of AI Generated Content

참조 기반 생성(예: 객체 합성, 커스터마이징)은 빠르게 발전해 왔으나, 기존 파이프라인들은 근본적인 한계를 공유합니다. 사용자가 제공한 객체 중심의 고해상도 참조 이미지(HRRI)가 모델에 입력되기 전에 고정된 저해상도(LR)로 다운샘플링되어, 출력이 생성되기도 전에 미세한 세부 정보가 버려진다는 점입니다. 게다가 생성 단계에서는 이러한 정보 손실 위에 고유한 아티팩트(예: 정체성 왜곡)가 추가로 발생합니다. 기존의 참조 기반 생성 콘텐츠 정제(RefGCR) 방식은 이러한 아티팩트 중 일부를 보정할 수 있지만 여전히 저해상도 영역에서 작동하며, 참조 기반 초해상화(RefSR) 방식은 해상도를 복원하지만 자연 이미지 열화를 가정하여 생성 파이프라인의 아티팩트 분포를 무시합니다. 이러한 두 가지 한계를 단일 프레임워크에서 해결하기 위해, 본 논문에서는 원본 HRRI를 후처리 단계에서 재사용하여 손실된 세부 정보를 복원하고, 생성 아티팩트를 정제하며, 출력을 동시에 업스케일링하는 새로운 태스크인 참조 기반 생성 콘텐츠 초해상화-정제(RefGC-SR$^2$)를 제안합니다. 우리는 이 RefGC-SR$^2$ 태스크를 위해 최초의 실제 삼중항(triplet) 데이터 생성 파이프라인을 구축하고, 공개된 사전학습 모델이 제공할 수 없는 쌍을 이루는 저품질 앵커를 합성하도록 딥틱 조건부 생성기를 학습시킵니다. 나아가 생성 아티팩트를 제거하면서 HRRI의 미세 세부 정보를 선택적으로 주입하는 RefGC-SR$^2$용 주파수 인식 확산 트랜스포머(diffusion transformer) 모델을 제시합니다. 광범위한 실험을 통해 제안하는 RefGC-SR$^2$ 모델이 (i) 참조 이미지에 충실하게 객체 정체성을 정제하고, (ii) 고해상도 세부 정보를 성공적으로 복원하여, 기존 RefGCR 및 RefSR 베이스라인과 비교해 최종 결과물이 현저히 높은 품질과 실질적인 활용성을 보임을 입증합니다.