NMNancy Mehta

Nancy Mehta

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Nancy Mehta has a total of 55 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Experts-Guided Unbalanced Optimal Transport for ISP Learning from Unpaired and/or Paired Data

    by Georgy Perevozchikov, Nancy Mehta, Egor Ershov, Radu Timofte

  • Open Access

    AIM 2025 Challenge on High FPS Motion Deblurring: Methods and Results

    by George Ciubotariu, Florin-Alexandru Vasluianu, Zhuyun Zhou, Nancy Mehta, Radu Timofte, Ke Wu, Long Sun, Lingshun Kong, Zhongbao Yang, Jinshan Pan, et al.

    This paper presents a comprehensive review of the AIM 2025 High FPS Non-Uniform Motion Deblurring Challenge, highlighting the proposed solutions and final results. The objective of this challenge is to identify effective networks capable of producing clearer and visually compelling images in diverse and challenging conditions, by learning representative visual cues for complex aggregations of motion types. A total of 68 participants registered for the competition, and 9 teams ultimately submitted valid entries. This paper thoroughly evaluates the state-of-the-art advances in high-FPS single image motion deblurring, showcasing the significant progress in the field, while leveraging samples of the novel dataset, MIORe, that introduces challenging examples of movement patterns.

  • Open Access

    AdaptSR: Low-Rank Adaptation for Efficient and Scalable Real-World Super-Resolution

    by Cansu Korkmaz, Nancy Mehta, Radu Timofte

    Recovering high-frequency details and textures from low-resolution images remains a fundamental challenge in super-resolution (SR), especially when real-world degradations are complex and unknown. While GAN-based methods enhance realism, they suffer from training instability and introduce unnatural artifacts. Diffusion models, though promising, demand excessive computational resources, often requiring multiple GPU days, even for single-step variants. Rather than naively fine-tuning entire models or adopting unstable generative approaches, we introduce AdaptSR, a low-rank adaptation (LoRA) framework that efficiently repurposes bicubic-trained SR models for real-world tasks. AdaptSR leverages architecture-specific insights and selective layer updates to optimize real SR adaptation. By updating only lightweight LoRA layers while keeping the pretrained backbone intact, it captures domain-specific adjustments without adding inference cost, as the adapted layers merge seamlessly post-training. This efficient adaptation not only reduces memory and compute requirements but also makes real-world SR feasible on lightweight hardware. Our experiments demonstrate that AdaptSR outperforms GAN and diffusion-based SR methods by up to 4 dB in PSNR and 2% in perceptual scores on real SR benchmarks. More impressively, it matches or exceeds full model fine-tuning while training 92% fewer parameters, enabling rapid adaptation to real SR tasks within minutes.

  • Probing Attention-Driven Normalizing Flow Network for Low-Light Image Enhancement

    by Siddharth Singh, Nancy Mehta, K. N. Prakash, Santosh Kumar Vipparthi, Subrahmanyam Murala

  • Rawformer: Unpaired Raw-to-Raw Translation for Learnable Camera ISPs

    by Georgy Perevozchikov, Nancy Mehta, Mahmoud Afifi, Radu Timofte


  • Image Super-Resolution With Content-Aware Feature Processing

    by Nancy Mehta, Subrahmanyam Murala

    Image super-resolution (SR) is currently a very active research topic with applications spanning from computer vision to videos and graphic industries. The top performers in SR field usually employ deep or wide convolutional neural networks (CNNs) to restore the lost textures from low-resolution images. However, most of these methods adopt pixel-shuffle, and deconvolution as their up-sampling techniques and often generate conspicuous artifacts in the reconstructed image. Additionally, the ongoing trend of directly portraying the degraded low-resolution image to high-resolution via complex deep CNNs improves the reconstruction performance, but at the cost of high computational complexity. In this work, we propose a multi-level bi-cubic up-sampler network (MBUp-Net) for reconstructing high-quality SR image with restricted number of parameters. A novel Content-Aware Feature Difference (CAFD) block is presented to reform the network by focusing on contextual information. The proposed CAFD block consists of four multi-level attention blocks for better extraction of low-level features at different scales. Furthermore, we design an innovative up-sampling layer that consistently outperforms the traditional up-sampling methods. These components collaboratively endow our proposed network with a great performance boost, helping it achieve state-of-the-art accuracy on five synthetic benchmark datasets, both qualitatively and quantitatively. In addition, a detailed ablation study has been accomplished to scrutinize the improvements obtained by different modules in the proposed method.


  • Con-Net: A Consolidated Light-Weight Image Restoration Network

    by Nancy Mehta, Subrahmanyam Murala

    There has been a considerable gap between the recent high-resolution display technologies and the short storage of its content. However, most of the existing restoration methods are restricted by local convolution operations and equal treatment of the diverse information in degraded image. These approaches being degradation-specific employ the same rigid spatial processing across different images ultimately resulting in high memory consumption. For overcoming this limitation we propose Con-Net, a network design capable of exploiting the non-uniformities of the degradations in spatial-domain with limited number of parameters (656k). Our proposed Con-Net comprises of basically two main components, (1) a spatial-degradation aware network for extracting the diverse information inherent in any degraded image, and (2) a holistic attention refinement network for exploiting the knowledge from the degradation aware network to selectively restore the degraded pixels. In a nutshell, our proposed method is generalizable for three applications: image denoising, super-resolution and real-world low-light enhancement. Extensive qualitative and quantitative comparison with prior arts on 8 benchmark datasets demonstrates the efficacy of our proposed Con-Net over existing state-of-the-art degradation-specific architectures, by huge parameter and FLOPs reduction in all the three tasks.


  • (MLE$^{2}$A$^{2}$U)-Net: Image Super-Resolution via Multi-Level Edge Embedding and Aggregated Attentive Upsampler Network

    by Nancy Mehta, Subrahmanyam Murala

    Given a degraded low-resolution input image, super-resolution (SR) aims at restoring the lost textures and structures and generating high-resolution image content. Significant advances in image super-resolution have been made lately, dominated by convolutional neural networks (CNNs). The top performing CNN-based SR networks typically employ very deep models for embracing the benefits of generating spatially precise results, but at the cost of loss of long-term contextual information. Additionally, state-of-the-art (SOTA) methods generally lack in maintaining the balance between spatial details and contextual information, which is the basic requirement for exhibiting superior performance in SR task. For restoration application like SR, the overall network generally demands efficient preservation of low-frequency information and reconstruction of high-frequency details. Thus, our work presents a novel architecture with the holistic objective of maintaining spatially-precise representation by collecting contextual content and restoring multi-frequency information throughout the network. Our proposed model learns an enriched set of features, that besides combining contextual information from multiple scales simultaneously preserves the high-resolution spatial details. The core of our approach is a novel non-local and local attention (NLLA) block which focuses on (1) learning enriched features by collecting information from multiple scales, (2) simultaneously handling the different frequency information, and (3) effectively fusing the relevant low-frequency and high-frequency information by ignoring the redundant features. Additionally, for effectively mapping the low-resolution features to high-resolution, we propose a novel aggregated attentive up-sampler (AAU) block that attentively learns the weights to up-sample the refined low-resolution feature maps to high-resolution output. Extensive experiments on the benchmark SR datasets demonstrate that the proposed method achieves appealing performance, both qualitatively and quantitatively.