Marwan Torki

Youssef Aboelwafa, Hicham G. Elmongui, Marwan Torki

Low-light image enhancement is challenging due to complex degradations, including amplified noise, artifacts, and color distortion. While Retinex-based deep learning methods have achieved promising results, they primarily rely on single-modality RGB information. We propose M2Retinexformer (Multi-Modal Retinexformer), a novel framework that extends Retinexformer by incorporating depth cues, luminance priors, and semantic features within a progressive refinement pipeline. Depth provides geometric context that is invariant to lighting variations, while luminance and semantic features offer explicit guidance on brightness distribution and scene understanding. Modalities are extracted at multiple scales and fused through cross-attention, with adaptive gating dynamically balancing illumination-guided self-attention and cross-attention based on the reliability of auxiliary cues. Evaluations on the LOL, SID, SMID, and SDSD benchmarks demonstrate overall improvements over Retinexformer and recent state-of-the-art methods. Code and pretrained weights are available at https://github.com/YoussefAboelwafa/M2Retinexformer

Ahmad Shapiro, Ayman Khalafallah, Marwan Torki · gatech

Online presence on social media platforms such as Facebook and Twitter has become a daily habit for internet users. Despite the vast amount of services the platforms offer for their users, users suffer from cyber-bullying, which further leads to mental abuse and may escalate to cause physical harm to individuals or targeted groups. In this paper, we present our submission to the Arabic Hate Speech 2022 Shared Task Workshop (OSACT5 2022) using the associated Arabic Twitter dataset. The shared task consists of 3 sub-tasks, sub-task A focuses on detecting whether the tweet is offensive or not. Then, For offensive Tweets, sub-task B focuses on detecting whether the tweet is hate speech or not. Finally, For hate speech Tweets, sub-task C focuses on detecting the fine-grained type of hate speech among six different classes. Transformer models proved their efficiency in classification tasks, but with the problem of over-fitting when fine-tuned on a small or an imbalanced dataset. We overcome this limitation by investigating multiple training paradigms such as Contrastive learning and Multi-task learning along with Classification fine-tuning and an ensemble of our top 5 performers. Our proposed solution achieved 0.841, 0.817, and 0.476 macro F1-average in sub-tasks A, B, and C respectively.

Yousef Kotp, Marwan Torki

Lens flare is a common image artifact that can significantly degrade image quality and affect the performance of computer vision systems due to a strong light source pointing at the camera. This survey provides a comprehensive overview of the multifaceted domain of lens flare, encompassing its underlying physics, influencing factors, types, and characteristics. It delves into the complex optics of flare formation, arising from factors like internal reflection, scattering, diffraction, and dispersion within the camera lens system. The diverse categories of flare are explored, including scattering, reflective, glare, orb, and starburst types. Key properties such as shape, color, and localization are analyzed. The numerous factors impacting flare appearance are discussed, spanning light source attributes, lens features, camera settings, and scene content. The survey extensively covers the wide range of methods proposed for flare removal, including hardware optimization strategies, classical image processing techniques, and learning-based methods using deep learning. It not only describes pioneering flare datasets created for training and evaluation purposes but also how they were created. Commonly employed performance metrics such as PSNR, SSIM, and LPIPS are explored. Challenges posed by flare's complex and data-dependent characteristics are highlighted. The survey provides insights into best practices, limitations, and promising future directions for flare removal research. Reviewing the state-of-the-art enables an in-depth understanding of the inherent complexities of the flare phenomenon and the capabilities of existing solutions. This can inform and inspire new innovations for handling lens flare artifacts and improving visual quality across various applications.

Abdelrahman Abouelenein, Marwan Torki

It is crucial for modern on-device AI systems that rely on retrieval-augmented inference to release and share datastores without compromising individual privacy. This can be achieved using Differential Privacy (DP), which provides a formal guarantee that ensures individual contributions remain indistinguishable, even under adversarial analysis. In this paper, we introduce a hashing-based probability generation framework designed to enable the creation and release of differentially private datastores. Our approach employs locality-sensitive hashing (LSH) to efficiently partition high-dimensional data into buckets. We then add calibrated DP noise to the accumulated vote for each bucket, generating a probability distribution across classes. Our method is broadly applicable to any pipeline requiring secure key,value datastore creation and release. We conducted experiments on seven datasets with varying sample sizes and class counts, ranging from 2 to 14. At epsilon=5, our released DP datastore achieves strong privacy protection with only an average 2.6% drop in accuracy. Finally, we benchmark DP datastore resilience to membership inference attacks, reducing attack accuracy to 53.60%.

Nour Shaheen, Tamer Elsharnouby, Marwan Torki

In scientific research, charts are usually the primary method for visually representing data. However, the accessibility of charts remains a significant concern. In an effort to improve chart understanding pipelines, we focus on optimizing the chart classification component. We leverage curriculum learning, which is inspired by the human learning process. In this paper, we introduce a novel training approach for chart classification that utilizes coarse-to-fine curriculum learning. Our approach, which we name C2F-CHART (for coarse-to-fine) exploits inter-class similarities to create learning tasks of varying difficulty levels. We benchmark our method on the ICPR 2022 CHART-Infographics UB UNITEC PMC dataset, outperforming the state-of-the-art results.

Nada Osman, Marwan Torki

Anomaly action detection and localization play an essential role in security and advanced surveillance systems. However, due to the tremendous amount of surveillance videos, most of the available data for the task is unlabeled or semi-labeled with the video class known, but the location of the anomaly event is unknown. In this work, we target anomaly localization in semi-supervised videos. While the mainstream direction in addressing this task is focused on segment-level multi-instance learning and the generation of pseudo labels, we aim to explore a promising yet unfulfilled direction to solve the problem by learning the temporal relations within videos in order to locate anomaly events. To this end, we propose a hierarchical transformer model designed to evaluate the significance of observed actions in anomalous videos with a divide-and-conquer strategy along the temporal axis. Our approach segments a parent video hierarchically into multiple temporal children instances and measures the influence of the children nodes in classifying the abnormality of the parent video. Evaluating our model on two well-known anomaly detection datasets, UCF-crime and ShanghaiTech, proves its ability to interpret the observed actions within videos and localize the anomalous ones. Our proposed approach outperforms previous works relying on segment-level multiple-instance learning approaches while reaching a promising performance compared to the more recent pseudo-labeling-based approaches.

Mariam Saeed, Manar Amr, Farida Adel et al.

Direct speech-to-image generation has recently shown promising results. However, compared to text-to-image generation, there is still a large gap to enclose. Current approaches use two stages to tackle this task: speech encoding network and image generative adversarial network (GAN). The speech encoding networks in these approaches produce embeddings that do not capture sufficient linguistic information to semantically represent the input speech. GANs suffer from issues such as non-convergence, mode collapse, and diminished gradient, which result in unstable model parameters, limited sample diversity, and ineffective generator learning, respectively. To address these weaknesses, we introduce a framework called Speak the Art (STA) which consists of a speech encoding network and a VQ-Diffusion network conditioned on speech embeddings. To improve speech embeddings, the speech encoding network is supervised by a large pre-trained image-text model during training. Replacing GANs with diffusion leads to more stable training and the generation of diverse images. Additionally, we investigate the feasibility of extending our framework to be multilingual. As a proof of concept, we trained our framework with two languages: English and Arabic. Finally, we show that our results surpass state-of-the-art models by a large margin.

Youssef Aboelwafa, Ahmed Samir, Nagwa Elmakky et al.

We present DistilledGemma, an efficient and accurate system for the HIPE-2026 shared task on person-place relation extraction from multilingual historical newspaper articles in English, German, and French. Our approach adopts a three-stage knowledge distillation pipeline designed to balance classification accuracy with computational efficiency. In the first stage, we systematically explored prompt engineering strategies across eight large language models to identify the most effective reasoning architecture for this challenging task. In the second stage, we applied supervised fine-tuning (SFT) via QLoRA to a Gemma 4 26B A4B teacher model, leveraging its strong multilingual capabilities to generate silver-standard chain-of-thought traces across the training corpus. In the final stage, we performed response-level distillation to transfer these learned reasoning patterns into a compact Gemma 4 E2B student model. In the official evaluation, our team WHEREAMI ranked 3rd on the standard test set with an accuracy profile mean score of 0.688, and 2nd on the binary test set with a mean score of 0.8156. Notably, by distilling knowledge from the 26B teacher to the 2.3B student, we preserved strong reasoning capabilities while reducing the deployed model size to approximately 2.3B effective parameters; the LoRA adapters used during training were merged into the student for inference. This configuration ranked 2nd in the balanced efficiency-accuracy profile across both the standard and binary test sets. These results demonstrate that knowledge distillation provides a practical and scalable solution for historical document processing, achieving competitive performance without excessive computational cost.

Abdelrahman Elnenaey, Marwan Torki

Image reflection removal is crucial for restoring image quality. Distorted images can negatively impact tasks like object detection and image segmentation. In this paper, we present a novel approach for image reflection removal using a single image. Instead of focusing on model architecture, we introduce a new training technique that can be generalized to image-to-image problems, with input and output being similar in nature. This technique is embodied in our multi-step loss mechanism, which has proven effective in the reflection removal task. Additionally, we address the scarcity of reflection removal training data by synthesizing a high-quality, non-linear synthetic dataset called RefGAN using Pix2Pix GAN. This dataset significantly enhances the model's ability to learn better patterns for reflection removal. We also utilize a ranged depth map, extracted from the depth estimation of the ambient image, as an auxiliary feature, leveraging its property of lacking depth estimations for reflections. Our approach demonstrates superior performance on the SIR^2 benchmark and other real-world datasets, proving its effectiveness by outperforming other state-of-the-art models.

Youssof Nawar, Nouran Soliman, Moustafa Wassel et al.

Glaucoma is a progressive optic neuropathy characterized by structural damage to the optic nerve head and functional changes in the visual field. Detecting glaucoma early is crucial to preventing loss of eyesight. However, medical datasets often suffer from class imbalances, making detection more difficult for deep-learning algorithms. We use a generative-based framework to enhance glaucoma diagnosis, specifically addressing class imbalance through synthetic data generation. In addition, we collected the largest national dataset for glaucoma detection to support our study. The imbalance between normal and glaucomatous cases leads to performance degradation of classifier models. By combining our proposed framework leveraging diffusion models with a pretraining approach, we created a more robust classifier training process. This training process results in a better-performing classifier. The proposed approach shows promising results in improving the harmonic mean (sensitivity and specificity) and AUC for the roc for the glaucoma classifier. We report an improvement in the harmonic mean metric from 89.09% to 92.59% on the test set of our national dataset. We examine our method against other methods to overcome imbalance through extensive experiments. We report similar improvements on the AIROGS dataset. This study highlights that diffusion-based generation can be of great importance in tackling class imbalances in medical datasets to improve diagnostic performance.

Ahmed Taha, Marwan Torki

In this paper, we cast the scribble-based interactive image segmentation as a semi-supervised learning problem. Our novel approach alleviates the need to solve an expensive generalized eigenvector problem by approximating the eigenvectors using efficiently computed eigenfunctions. The smoothness operator defined on feature densities at the limit n tends to infinity recovers the exact eigenvectors of the graph Laplacian, where n is the number of nodes in the graph. To further reduce the computational complexity without scarifying our accuracy, we select pivots pixels from user annotations. In our experiments, we evaluate our approach using both human scribble and "robot user" annotations to guide the foreground/background segmentation. We developed a new unbiased collection of five annotated images datasets to standardize the evaluation procedure for any scribble-based segmentation method. We experimented with several variations, including different feature vectors, pivot count and the number of eigenvectors. Experiments are carried out on datasets that contain a wide variety of natural images. We achieve better qualitative and quantitative results compared to state-of-the-art interactive segmentation algorithms.

Moustafa Meshry, Mohamed E. Hussein, Marwan Torki

Sliding window is one direct way to extend a successful recognition system to handle the more challenging detection problem. While action recognition decides only whether or not an action is present in a pre-segmented video sequence, action detection identifies the time interval where the action occurred in an unsegmented video stream. Sliding window approaches for action detection can however be slow as they maximize a classifier score over all possible sub-intervals. Even though new schemes utilize dynamic programming to speed up the search for the optimal sub-interval, they require offline processing on the whole video sequence. In this paper, we propose a novel approach for online action detection based on 3D skeleton sequences extracted from depth data. It identifies the sub-interval with the maximum classifier score in linear time. Furthermore, it is invariant to temporal scale variations and is suitable for real-time applications with low latency.

Mohamed E. Hussein, Marwan Torki, Ahmed Elsallamy et al.

In this paper, we introduce the first phase of a new dataset for offline Arabic handwriting recognition. The aim is to collect a very large dataset of isolated Arabic words that covers all letters of the alphabet in all possible shapes using a small number of simple words. The end goal is to collect a very large dataset of segmented letter images, which can be used to build and evaluate Arabic handwriting recognition systems that are based on segmented letter recognition. The current version of the dataset contains $25114$ samples of $109$ unique Arabic words that cover all possible shapes of all alphabet letters. The samples were collected from $907$ writers. In its current form, the dataset can be used for the problem of closed-vocabulary word recognition. We evaluated a number of window-based descriptors and classifiers on this task and obtained an accuracy of $92.16\%$ using a SIFT-based descriptor and ANN.

Marwan Torki, Mohamed E. Hussein, Ahmed Elsallamy et al.

This paper presents a comparative study for window-based descriptors on the application of Arabic handwritten alphabet recognition. We show a detailed experimental evaluation of different descriptors with several classifiers. The objective of the paper is to evaluate different window-based descriptors on the problem of Arabic letter recognition. Our experiments clearly show that they perform very well. Moreover, we introduce a novel spatial pyramid partitioning scheme that enhances the recognition accuracy for most descriptors. In addition, we introduce a novel dataset for Arabic handwritten isolated alphabet letters, which can serve as a benchmark for future research.