arXiv:2512.05021v1 Announce Type: new 
Abstract: Handwritten Text Recognition remains challenging due to the limited data, high writing style variance, and scripts with complex diacritics. Existing approaches, though partially address these issues, often struggle to generalize without massive synthetic data. To address these challenges, we propose HTR-ConvText, a model designed to capture fine-grained, stroke-level local features while preserving global contextual dependencies. In the feature extraction stage, we integrate a residual Convolutional Neural Network backbone with a MobileViT with Positional Encoding block. This enables the model to both capture structural patterns and learn subtle writing details. We then introduce the ConvText encoder, a hybrid architecture combining global context and local features within a hierarchical structure that reduces sequence length for improved efficiency. Additionally, an auxiliary module injects textual context to mitigate the weakness of Connectionist Temporal Classification. Evaluations on IAM, READ2016, LAM and HANDS-VNOnDB demonstrate that our approach achieves improved performance and better generalization compared to existing methods, especially in scenarios with limited training samples and high handwriting diversity.

تم تقديم نموذج جديد يسمى HTR-ConvText لتحسين التعرف على النصوص المكتوبة بخط اليد، حيث يتناول التحديات مثل نقص البيانات وتنوع أنماط الكتابة. يدمج هذا النموذج شبكة عصبية تلافيفية متبقية مع بنية MobileViT، مما يمكّنه من التقاط الميزات المحلية والسياق العالمي بشكل فعال.

Se ha presentado un nuevo modelo llamado HTR-ConvText para mejorar el reconocimiento de texto manuscrito, abordando desafíos como la escasez de datos y la alta variabilidad en los estilos de escritura. Este modelo integra una red neuronal convolucional residual con una arquitectura MobileViT, lo que le permite capturar de manera efectiva tanto características locales como contexto global.

Un nouveau modèle nommé HTR-ConvText a été introduit pour améliorer la reconnaissance de texte manuscrit, en s'attaquant à des défis tels que la quantité limitée de données et la grande variance dans les styles d'écriture. Ce modèle intègre un réseau de neurones convolutifs résiduels avec une architecture MobileViT, lui permettant de capturer efficacement à la fois les caractéristiques locales et le contexte global.

A new model named HTR-ConvText has been introduced to enhance handwritten text recognition, addressing challenges such as limited data and high variance in writing styles. This model integrates a residual Convolutional Neural Network with a MobileViT architecture, enabling it to capture both local features and global context effectively.

HTR-ConvText: Leveraging Convolution and Textual Information for Handwritten Text Recognition

arXiv:2512.04464v1 Announce Type: new 
Abstract: We challenge the common belief that deep learning always trumps older techniques, using the example of grading Saint-Gaudens Double Eagle gold coins automatically. In our work, we put a feature-based Artificial Neural Network built around 192 custom features pulled from Sobel edge detection and HSV color analysis up against a hybrid Convolutional Neural Network that blends in EfficientNetV2, plus a straightforward Support Vector Machine as the control. Testing 1,785 coins graded by experts, the ANN nailed 86% exact matches and hit 98% when allowing a 3-grade leeway. On the flip side, CNN and SVM mostly just guessed the most common grade, scraping by with 31% and 30% exact hits. Sure, the CNN looked good on broader tolerance metrics, but that is because of some averaging trick in regression that hides how it totally flops at picking out specific grades. All told, when you are stuck with under 2,000 examples and lopsided classes, baking in real coin-expert knowledge through feature design beats out those inscrutable, all-in-one deep learning setups. This rings true for other niche quality checks where data's thin and know-how matters more than raw compute.

تم إجراء دراسة مقارنة حول تصنيف العملات الذهبية من نوع سانت غودنز دابل إيجل تلقائيًا، مما يتحدى الفكرة القائلة بأن التعلم العميق يتفوق دائمًا على الأساليب التقليدية. اختبرت الدراسة شبكة عصبية اصطناعية (ANN) تستخدم 192 ميزة مخصصة مقابل شبكة عصبية تلافيفية (CNN) هجينة وآلة دعم المتجهات (SVM)، وكشفت أن ANN حققت 86% من التطابقات الدقيقة مقارنة بـ 31% لـ CNN و30% لـ SVM.

Se ha realizado un estudio comparativo sobre la clasificación automatizada de monedas de oro Saint-Gaudens Double Eagle, desafiando la noción de que el aprendizaje profundo supera constantemente a los métodos tradicionales. El estudio probó una Red Neuronal Artificial (ANN) que utiliza 192 características personalizadas frente a una Red Neuronal Convolucional (CNN) híbrida y una Máquina de Vectores de Soporte (SVM), revelando que la ANN logró un 86% de coincidencias exactas en comparación con el 31% de la CNN y el 30% de la SVM.

Une étude comparative a été réalisée sur la notation automatisée des pièces d'or Saint-Gaudens Double Eagle, remettant en question l'idée selon laquelle l'apprentissage profond surpasse systématiquement les méthodes traditionnelles. L'étude a testé un réseau de neurones artificiels (ANN) utilisant 192 caractéristiques personnalisées contre un réseau de neurones convolutifs (CNN) hybride et une machine à vecteurs de support (SVM), révélant que l'ANN a atteint 86 % de correspondances exactes contre 31 % pour le CNN et 30 % pour le SVM.

A comparative study has been conducted on automated grading of Saint-Gaudens Double Eagle gold coins, challenging the notion that deep learning consistently outperforms traditional methods. The study tested an Artificial Neural Network (ANN) utilizing 192 custom features against a hybrid Convolutional Neural Network (CNN) and a Support Vector Machine (SVM), revealing that the ANN achieved 86% exact matches compared to the CNN's 31% and SVM's 30%.

Feature Engineering vs. Deep Learning for Automated Coin Grading: A Comparative Study on Saint-Gaudens Double Eagles

arXiv:2506.14987v3 Announce Type: replace-cross 
Abstract: Ensuring packet-level communication quality is vital for ultra-reliable, low-latency communications (URLLC) in large-scale industrial wireless networks. We enhance the Local Deadline Partition (LDP) algorithm by introducing a CNN-based dynamic priority prediction mechanism for improved interference coordination in multi-cell, multi-channel networks. Unlike LDP's static priorities, our approach uses a Convolutional Neural Network and graph coloring to adaptively assign link priorities based on real-time traffic, transmission opportunities, and network conditions. Assuming that first training phase is performed offline, our approach introduced minimal overhead, while enabling more efficient resource allocation, boosting network capacity, SINR, and schedulability. Simulation results show SINR gains of up to 113\%, 94\%, and 49\% over LDP across three network configurations, highlighting its effectiveness for complex URLLC scenarios.

تم تقديم نهج جديد للجدولة في الاتصالات فائقة الموثوقية ومنخفضة الكمون (URLLC) من خلال تحسين خوارزمية تقسيم المواعيد المحلي (LDP) باستخدام آلية توقع أولوية ديناميكية تعتمد على الشبكات العصبية التلافيفية (CNN). تهدف هذه الابتكارات إلى تحسين تنسيق التداخل في الشبكات المعقدة متعددة الخلايا والقنوات من خلال تخصيص أولويات الروابط بشكل تكيفي بناءً على الظروف في الوقت الحقيقي.

Se ha introducido un nuevo enfoque para la programación en comunicaciones ultra confiables y de baja latencia (URLLC), mejorando el algoritmo de Local Deadline Partition (LDP) con un mecanismo de predicción de prioridad dinámica basado en CNN. Esta innovación tiene como objetivo mejorar la coordinación de interferencias en redes complejas de múltiples celdas y múltiples canales al asignar de manera adaptativa prioridades de enlace según las condiciones en tiempo real.

Une nouvelle approche de planification dans les communications ultra-fiables et à faible latence (URLLC) a été introduite, améliorant l'algorithme de Local Deadline Partition (LDP) avec un mécanisme de prédiction de priorité dynamique basé sur CNN. Cette innovation vise à améliorer la coordination des interférences dans des réseaux complexes multi-cellules et multi-canaux en attribuant de manière adaptative des priorités de lien en fonction des conditions en temps réel.

A new approach to scheduling in ultra-reliable, low-latency communications (URLLC) has been introduced, enhancing the Local Deadline Partition (LDP) algorithm with a CNN-based dynamic priority prediction mechanism. This innovation aims to improve interference coordination in complex multi-cell, multi-channel networks by adaptively assigning link priorities based on real-time conditions.

HTR-ConvText: Leveraging Convolution and Textual Information for Handwritten Text Recognition

Was this article worth reading? Share it

Humanize AI

Airparser

AI Humanizer