VANET Lab IIT Jodhpur

Our recent work, “Addressing Long-Tailed Spatial and Category Imbalances in Citywide Incident Prediction", accepted in 𝐈𝐄𝐄𝐄 𝐓𝐫𝐚𝐧𝐬𝐚𝐜𝐭𝐢𝐨𝐧𝐬 𝐨𝐧 𝐁𝐢𝐠 𝐃𝐚𝐭𝐚 (𝐈𝐦𝐩𝐚𝐜𝐭 𝐅𝐚𝐜𝐭𝐨𝐫: 𝟓.𝟕), focuses on improving citywide incident prediction under highly imbalanced real-world urban data. In smart city environments, incident records are often unevenly distributed across both locations and incident categories, where a few regions and frequent events dominate the dataset while many critical but rare incidents remain under-represented. This long-tailed nature makes conventional prediction models biased toward majority patterns and less effective for minority regions and rare incident types.

🌐 𝐄𝐚𝐫𝐥𝐲 𝐚𝐜𝐜𝐞𝐬𝐬 𝐥𝐢𝐧𝐤: https://ieeexplore.ieee.org/document/11460224

𝐂𝐨𝐧𝐠𝐫𝐚𝐭𝐮𝐥𝐚𝐭𝐢𝐨𝐧𝐬 𝐭𝐨 𝐚𝐥𝐥 𝐭𝐡𝐞 𝐚𝐮𝐭𝐡𝐨𝐫𝐬:
Bhumika Chaudhary, PhD student at Department of Computer Science and Engineering, IIT Jodhpur
Dr. Debasis Das, Associate Professor at Department of Computer Science and Engineering, IIT Jodhpur

𝐒𝐮𝐦𝐦𝐚𝐫𝐲:
Citywide incidents such as crimes, accidents, and public safety threats contribute to substantial societal disruption and economic loss. Accurate prediction of such incidents can significantly aid city administrators in proactive response planning. Existing approaches model the incident prediction as a spatio-temporal task but often neglect the inter-region spatial long-tailed distribution of incidents. This uneven distribution introduces spatial bias in learning, which causes models to overfit regions with frequent incidents (head regions) while underfitting the regions with occasional incidents (tail regions). Furthermore, model learning is hindered by intra-region category imbalance, where certain incident types (e.g., theft) dominate over rarer categories (e.g., robbery) within the same region. To address inter- and intra-region challenges, we propose an approach named SLIP (Spatial Long-tail Incident Prediction). Specifically, for inter-region skewness, SLIP adopts a multi-expert design comprising a common feature extraction backbone followed by three expert branches. In addition, to mitigate the intra-region category imbalance, we utilizes a variant of focal loss, particularly for positive-negative imbalance. SLIP outperforms spatio-temporal state-of-the-art methods by 2-11% in Macro F1, 4-11% in Micro F1, and 1-11% in Severity Weighted F1 across Los Angeles and Chicago cities for the urban crime dataset. Additionally, we incorporate fairness metrics into the evaluation and present a comprehensive comparison of spatio-temporal incident prediction.

Our work titled "𝐄𝐏𝐔𝐕: 𝐀 𝐋𝐢𝐠𝐡𝐭𝐰𝐞𝐢𝐠𝐡𝐭 𝐏𝐫𝐨𝐭𝐨𝐜𝐨𝐥 𝐟𝐨𝐫 𝐄𝐧𝐡𝐚𝐧𝐜𝐢𝐧𝐠 𝐒𝐞𝐜𝐮𝐫𝐢𝐭𝐲 𝐚𝐧𝐝 𝐏𝐞𝐫𝐟𝐨𝐫𝐦𝐚𝐧𝐜𝐞 𝐢𝐧 𝐔𝐀𝐕-𝐀𝐬𝐬𝐢𝐬𝐭𝐞𝐝 𝐈𝐨𝐕 𝐒𝐲𝐬𝐭𝐞𝐦𝐬" has been published at 𝐄𝐥𝐬𝐞𝐯𝐢𝐞𝐫 𝐏𝐞𝐫𝐯𝐚𝐬𝐢𝐯𝐞 𝐚𝐧𝐝 𝐌𝐨𝐛𝐢𝐥𝐞 𝐂𝐨𝐦𝐩𝐮𝐭𝐢𝐧𝐠 [𝐈𝐦𝐩𝐚𝐜𝐭 𝐅𝐚𝐜𝐭𝐨𝐫: 𝟑.𝟓].

🌐𝐂𝐡𝐞𝐜𝐤 𝐢𝐭 𝐡𝐞𝐫𝐞: https://www.sciencedirect.com/science/article/pii/S1574119226000477

Congratulations to all the authors:
Haradhan Ghosh, Research Associate at IIT Kharagpur
Ayanabha Ghosh, PhD student at
Dr. Debasis Das, Associate Professor at the Department of Computer Science and Engineering, IIT Jodhpur
Dr. Satya Bagchi, Associate Professor at the Department of Mathematics,

*Haradhan did this work as a Visiting Researcher at VANET Lab IIT Jodhpur.

𝐒𝐮𝐦𝐦𝐚𝐫𝐲:
The integration of UAVs into Internet of Vehicles (IoV) networks enhances coverage and responsiveness but also amplifies security and efficiency challenges. Existing IoV and UAV-assisted authentication schemes often rely on heavy cryptographic operations or complex multi-factor mechanisms, resulting in high computation and energy overhead unsuitable for resource-constrained UAVs. Many also lack complete mutual authentication across vehicle–UAV–RSU entities or fail to provide essential properties such as forward secrecy, resistance to impersonation and short-term secret leakage, and comprehensive formal verification. To address these gaps, we propose EPUV, a lightweight authentication and key-exchange protocol designed specifically for UAV-assisted IoV environments. EPUV employs only efficient primitives, elliptic curve cryptography (ECC), hashing, XOR, and concatenation, while providing strong anonymity, unlinkability, mutual authentication, and resistance to major attacks. Its security is rigorously validated through BAN logic, the Random Oracle Model, Scyther verification, and a Python-based symbolic verification framework, which confirms the correctness of protocol computations and resistance to guessing, replay, and impersonation attacks. Experimental evaluation on desktop and Raspberry Pi platforms shows that EPUV reduces computation overhead by 33.09% and energy consumption by 26.87% compared to state-of-the-art schemes, making it highly suitable for real-time, resource-constrained IoV applications.

𝐇𝐢𝐠𝐡𝐥𝐢𝐠𝐡𝐭𝐬:
- Designed to optimize computational efficiency while ensuring robust security properties.
- Python-based symbolic verification.
- Reduced computational overhead and energy consumption.

𝐅𝐞𝐝𝐂𝐫𝐢𝐦𝐞: 𝐙𝐞𝐫𝐨-𝐢𝐧𝐟𝐥𝐚𝐭𝐢𝐨𝐧 𝐀𝐝𝐚𝐩𝐭𝐢𝐯𝐞 𝐅𝐞𝐝𝐞𝐫𝐚𝐭𝐞𝐝 𝐋𝐞𝐚𝐫𝐧𝐢𝐧𝐠 𝐟𝐨𝐫 𝐂𝐫𝐢𝐦𝐞 𝐏𝐫𝐞𝐝𝐢𝐜𝐭𝐢𝐨𝐧 has been accepted and published at 𝐄𝐥𝐬𝐞𝐯𝐢𝐞𝐫 𝐍𝐞𝐮𝐫𝐨𝐂𝐨𝐦𝐩𝐮𝐭𝐢𝐧𝐠 [𝐈𝐦𝐩𝐚𝐜𝐭 𝐅𝐚𝐜𝐭𝐨𝐫: 𝟔.𝟓].

Congratulations to all the authors:
Bhumika Chaudhary, PhD student at the Department of Computer Science and Engineering, IIT Jodhpur
Prof. Philippe Lalanda, Université Grenoble Alpes
Dr. German Vega, Université Grenoble Alpes
Dr. Debasis Das, Associate Professor at the Department of Computer Science and Engineering, IIT Jodhpur

🌐Paper Link: https://www.sciencedirect.com/science/article/pii/S0925231226006144
🖥️Code: https://github.com/vanetlabiitj/FedCrime

𝐒𝐮𝐦𝐦𝐚𝐫𝐲:
In smart cities, crime prediction plays a critical role in enhancing public safety through proactive interventions and efficient resource allocation. In recent years, deep learning based approaches have shown promising performance for spatio-temporal crime forecasting. However, most existing methods rely on centralized training, where sensitive historical data are collected in a single repository, raising significant privacy concerns and limiting scalability. Federated learning offers a compelling alternative by enabling collaborative model training across multiple clients (e.g., police precincts) without sharing raw data. Nevertheless, conventional federated learning techniques are poorly suited to crime prediction, as crime data exhibit long-tailed spatial heterogeneity which causes zero inflation in low-frequency (tail) regions. In such settings, tail regions suffer from limited predictive performance, while their noisy updates can negatively affect global aggregation, leading to negative knowledge transfer. To address these challenges, we propose FedCrime, a federated learning framework specifically designed for crime prediction under extreme sparsity. FedCrime combines a temporal model at the client level with a sparsity aware zero-inflated loss function that explicitly accounts for excess zeros and over dispersed crime counts. This design aligns local optimization with the underlying data generation process, enabling stable and effective aggregation using standard Federated Averaging.

𝗧𝗵𝗶𝘀 𝗶𝘀 𝗮 𝗰𝗼𝗹𝗹𝗮𝗯𝗼𝗿𝗮𝘁𝗶𝘃𝗲 𝘄𝗼𝗿𝗸 𝗯𝗲𝘁𝘄𝗲𝗲𝗻 VANET Lab IIT Jodhpur 𝗮𝗻𝗱 𝗧𝗵𝗲 𝗗𝗲𝗽𝗮𝗿𝘁𝗺𝗲𝗻𝘁 𝗼𝗳 𝗖𝗼𝗺𝗽𝘂𝘁𝗲𝗿 𝗦𝗰𝗶𝗲𝗻𝗰𝗲, Université Grenoble Alpes (𝗨𝗚𝗔), 𝗙𝗿𝗮𝗻𝗰𝗲.

'𝐑𝐏𝐨𝐄: 𝐀 𝐑𝐞𝐩𝐮𝐭𝐚𝐭𝐢𝐨𝐧-𝐀𝐰𝐚𝐫𝐞 𝐏𝐫𝐨𝐨𝐟-𝐨𝐟-𝐄𝐧𝐞𝐫𝐠𝐲 𝐁𝐥𝐨𝐜𝐤𝐜𝐡𝐚𝐢𝐧 𝐏𝐫𝐨𝐭𝐨𝐜𝐨𝐥 𝐟𝐨𝐫 𝐄𝐥𝐞𝐜𝐭𝐫𝐢𝐜 𝐕𝐞𝐡𝐢𝐜𝐥𝐞 𝐄𝐧𝐞𝐫𝐠𝐲 𝐓𝐫𝐚𝐝𝐢𝐧𝐠 𝐢𝐧 𝐈𝐨𝐄𝐕' has been accepted at 𝐈𝐄𝐄𝐄 𝐓𝐫𝐚𝐧𝐬𝐚𝐜𝐭𝐢𝐨𝐧𝐬 𝐨𝐧 𝐕𝐞𝐡𝐢𝐜𝐮𝐥𝐚𝐫 𝐓𝐞𝐜𝐡𝐧𝐨𝐥𝐨𝐠𝐲 [𝐈𝐅: 𝟕.𝟏]

🌐 Early access link: https://ieeexplore.ieee.org/document/11408933

Congratulations to all the authors:
Koustav Kumar Mondal, PhD student at
Amritesh Kumar, Faculty member at Bennett University
Dr. Debasis Das, Associate Professor at Department of Computer Science and Engineering, IIT Jodhpur

𝐒𝐮𝐦𝐦𝐚𝐫𝐲:
The future of EV energy trading is decentralized, but are our current blockchain models efficient enough to handle city-scale networks?

As the Internet of Electric Vehicles (IoEV) grows, traditional centralized charging networks are increasingly held back by single points of failure, transactional opacity, and privacy concerns. While decentralized blockchain networks offer a promising alternative, standard consensus mechanisms like Proof-of-Work (PoW), Proof-of-Stake (PoS), and PBFT struggle with massive energy requirements, centralization risks, and heavy communication overheads that limit scalability. Enter 𝐑𝐏𝐨𝐄 (𝐑𝐞𝐩𝐮𝐭𝐚𝐭𝐢𝐨𝐧-𝐀𝐰𝐚𝐫𝐞 𝐏𝐫𝐨𝐨𝐟-𝐨𝐟-𝐄𝐧𝐞𝐫𝐠𝐲), a newly proposed, lightweight blockchain consensus protocol engineered specifically for peer-to-peer EV energy trading.

Here is how RPoE is changing the game for smart cities and the IoEV:
🔋 Tying Influence to Efficiency
⚡ Massive Performance Gains
⚖️ Fair and Balanced Incentives
🛡️ Robust Security

RPoE represents a major step forward in building an efficient, secure, and highly transparent energy-trading ecosystem for the vehicles of tomorrow.

'𝐪𝐈𝐨𝐕: 𝐀 𝐪𝐮𝐚𝐧𝐭𝐮𝐦-𝐝𝐫𝐢𝐯𝐞𝐧 𝐚𝐩𝐩𝐫𝐨𝐚𝐜𝐡 𝐟𝐨𝐫 𝐞𝐧𝐯𝐢𝐫𝐨𝐧𝐦𝐞𝐧𝐭𝐚𝐥 𝐦𝐨𝐧𝐢𝐭𝐨𝐫𝐢𝐧𝐠 𝐚𝐧𝐝 𝐫𝐚𝐩𝐢𝐝 𝐫𝐞𝐬𝐩𝐨𝐧𝐬𝐞 𝐬𝐲𝐬𝐭𝐞𝐦𝐬 𝐮𝐬𝐢𝐧𝐠 𝐢𝐧𝐭𝐞𝐫𝐧𝐞𝐭 𝐨𝐟 𝐯𝐞𝐡𝐢𝐜𝐥𝐞𝐬' has been accepted at Elsevier Ad-hoc Networks [IF: 4.8]

Congratulations to all the authors:

Dr. Er Ankur Nahar , PhD, Harish and Bina Shah School of AI and Computer Science, , India

Koustav Kumar Mondal, Centre for AIoT and Applications, , India

Dr. Debasis Das, Associate Professor, Department of Computer Science and Engineering, IIT Jodhpur, India

Prof. Rajkumar Buyya, Redmond Barry Distinguished Professor, School of Computing and Information Systems, The University of Melbourne, Australia.

𝐒𝐮𝐦𝐦𝐚𝐫𝐲 :
In today’s fast-paced world, rapid response to environmental threats such as urban gas leaks and industrial emissions is more critical than ever. However, traditional sensor networks often struggle with processing complex, high-dimensional environmental data in real time. To address this challenge, we introduce qIoV - an innovative framework that integrates Quantum Computing with the Internet of Vehicles (IoV) to redefine large-scale environmental monitoring.

Vehicles as Mobile Environmental Sensors
Imagine transforming everyday vehicles into intelligent environmental watchdogs. Equipped with advanced multi-gas sensors (e.g., MQ-series), vehicles can continuously monitor air quality, detect trace gases, and cover vast urban areas-identifying hazards that stationary systems may miss.

Variational Quantum Classifier (VQC)
At the core of qIoV lies a quantum machine learning model that maps complex sensor data into quantum states. The VQC captures deep correlations within high-dimensional data—patterns that classical algorithms often fail to recognize-enhancing detection accuracy and responsiveness.

Quantum Mesh Network Fabric (QMF)
In emergency scenarios, response time is everything. By leveraging principles such as quantum entanglement and teleportation, qIoV enables ultra-responsive communication across a dynamic vehicular network. As vehicles move, the network self-adapts, maintaining stability and delivering alerts with minimal latency.

qIoV marks a significant step toward smarter, safer, and more sustainable cities. By merging quantum intelligence with connected vehicle ecosystems, we envision a future where every vehicle actively contributes to environmental safety and urban resilience.

This is the collaborative work between VANET Lab IIT Jodhpur and 𝐓𝐡𝐞 𝐐𝐮𝐚𝐧𝐭𝐮𝐦 𝐂𝐥𝐨𝐮𝐝 𝐂𝐨𝐦𝐩𝐮𝐭𝐢𝐧𝐠 𝐚𝐧𝐝 𝐃𝐢𝐬𝐭𝐫𝐢𝐛𝐮𝐭𝐞𝐝 𝐒𝐲𝐬𝐭𝐞𝐦𝐬 (𝐪𝐂𝐋𝐎𝐔𝐃𝐒) 𝐋𝐚𝐛 at the University of Melbourne, Australia.

We are delighted to share that Ayanabha Ghosh, a third-year PhD student at VANET Lab IIT Jodhpur, working under the guidance of Dr. Debasis Das, has been selected as a 𝐏𝐡𝐃 𝐑𝐞𝐬𝐞𝐚𝐫𝐜𝐡 𝐈𝐧𝐭𝐞𝐫𝐧 (𝐀𝐈) at IBM 𝐑𝐞𝐬𝐞𝐚𝐫𝐜𝐡, Bengaluru.

His current research focuses on 𝐜𝐚𝐮𝐬𝐚𝐥 𝐝𝐢𝐬𝐜𝐨𝐯𝐞𝐫𝐲, 𝐜𝐚𝐮𝐬𝐚𝐥 𝐢𝐧𝐟𝐞𝐫𝐞𝐧𝐜𝐞, 𝐞𝐱𝐩𝐥𝐚𝐢𝐧𝐚𝐛𝐢𝐥𝐢𝐭𝐲, 𝐚𝐧𝐝 𝐭𝐫𝐮𝐬𝐭𝐰𝐨𝐫𝐭𝐡𝐲 𝐥𝐞𝐚𝐫𝐧𝐢𝐧𝐠 𝐦𝐞𝐜𝐡𝐚𝐧𝐢𝐬𝐦𝐬 for 𝐭𝐢𝐦𝐞-𝐝𝐲𝐧𝐚𝐦𝐢𝐜𝐚𝐥 𝐬𝐲𝐬𝐭𝐞𝐦𝐬. During this internship, he will be working with the amazing team behind state-of-the-art time series foundation models, including 𝐆𝐫𝐚𝐧𝐢𝐭𝐞 𝐓𝐓𝐌 and 𝐆𝐫𝐚𝐧𝐢𝐭𝐞 𝐓𝐒𝐏𝐮𝐥𝐬𝐞, contributing at the intersection of causality and foundation models.

We congratulate Ayanabha on this well-deserved opportunity and wish him continued success in his research endeavours.

We are ready to host prospective 𝙋𝙤𝙨𝙩𝙙𝙤𝙘𝙩𝙤𝙧𝙖𝙡 𝙁𝙚𝙡𝙡𝙤𝙬𝙨 in various research areas under the 2026 programme.

𝐎𝐟𝐟𝐢𝐜𝐢𝐚𝐥 𝐀𝐩𝐩𝐥𝐢𝐜𝐚𝐭𝐢𝐨𝐧 𝐛𝐞𝐠𝐢𝐧𝐬: 𝐉𝐚𝐧 𝟏𝟓, 𝟐𝟎𝟐𝟔
More details can be found here: https://anrfonline.in/ANRF/npdf

𝐈𝐧𝐭𝐞𝐫𝐞𝐬𝐭𝐞𝐝 𝐜𝐚𝐧𝐝𝐢𝐝𝐚𝐭𝐞𝐬 𝐚𝐫𝐞 𝐫𝐞𝐪𝐮𝐞𝐬𝐭𝐞𝐝 𝐭𝐨:
Fill up the interest form (must) - https://forms.gle/omfkH23f6NsYJH6X7
and contact Dr. Debasis Das (debasis[at]iitj.ac.in), Associate Professor at the Department of Computer Science and Engineering, IIT Jodhpur, to discuss further.

𝐑𝐞𝐬𝐞𝐚𝐫𝐜𝐡 𝐚𝐫𝐞𝐚𝐬:
- Explainable AI, Causal ML, GenAI, Diffusion Models
- LLMs, SLMs, VLMs, and other different aspects of Foundation Models
- Time Series Analysis, Multimodal data, Graph Neural Nets
- Edge intelligence, Federated learning, Lightweight/Resource constrained DL
- Autonomous Driving and Collaborative AVs
- Blockchain and Cybersecurity
- Hardware and network security
.. and their applications of Vehicular Ad-hoc Networks, Intelligent Transportation, Urban Intelligence, Vehicular Diagnostics and health management, Smart City, Smart Healthcare, and Smart Grid.

𝐋𝐚𝐛 𝐖𝐞𝐛𝐬𝐢𝐭𝐞: https://lnkd.in/dCDzDFJU

𝐖𝐡𝐚𝐭 𝐰𝐞 𝐨𝐟𝐟𝐞𝐫:
- State-of-the-art GPU computing facilities, including A100, A6000, and L40S.
- Real-time hardware testbeds.
- Advanced networking tools.

New acceptance at 𝐈𝐄𝐄𝐄 𝐓𝐫𝐚𝐧𝐬𝐚𝐜𝐭𝐢𝐨𝐧𝐬 𝐨𝐧 𝐂𝐨𝐧𝐬𝐮𝐦𝐞𝐫 𝐄𝐥𝐞𝐜𝐭𝐫𝐨𝐧𝐢𝐜𝐬. [IF: 10.9]

𝐓𝐢𝐭𝐥𝐞: "Secure IoT Communications with Optimized AES and Dynamic Threat Modeling on Embedded System"

𝐀𝐮𝐭𝐡𝐨𝐫𝐬:
Koustav Kumar Mondal, PhD student at Indian Institute of Technology Jodhpur
Dr. Himani Sikarwar, Assistant Professor of CSE, Shiv Nadar Institution of Eminence Deemed to be University Delhi-NCR
Dr. Debasis Das, Associate Professor, Department of Computer Science and Engineering, IIT Jodhpur
Prof. Chun-I Fan, Distinguished Professor, National Sun Yat-Sen University, Taiwan.

Congratulations to all the authors.

𝐄𝐚𝐫𝐥𝐲 𝐀𝐜𝐜𝐞𝐬𝐬 𝐋𝐢𝐧𝐤: https://ieeexplore.ieee.org/document/11301744

𝐒𝐮𝐦𝐦𝐚𝐫𝐲: This paper presents a hardware-algorithm co-design for secure IoT communication using an ESP32 microcontroller and MPU6050 sensor. The authors introduce a lightweight AES-128 optimization featuring a two-phase 4-bit S-box and simplified MixColumns that reduces encryption time by up to 42.4% and energy use by 28.57%. The architecture physically separates cipher and hash datapaths to enable true concurrency for protocols like TLS/IPsec. Security is bolstered by ephemeral ECDH for forward secrecy and a dynamic threat-modelling algorithm that quantifies risks across physical, side-channel, and network vectors. Testing confirms a high throughput of 435 Mbps and demonstrates that these optimizations are portable across other MCU platforms like Arm Cortex-M0+ and M33.

We are pleased to announce three new journal publications by VANET Lab IIT Jodhpur members. Congratulations to all authors for their impactful contributions.

1. DyBatch: Message Prioritization and Priority-driven Dynamic Batch Verification in Large-scale IoV Networks
in IEEE Transactions on Vehicular Technology, 2025 [IF: 7.1]
Authors: Himani Sikarwar, Ankur Nahar, Debasis Das
Link: https://ieeexplore.ieee.org/document/11203265

This work introduces DyBatch, a dynamic verification model that prioritizes alert messages and adapts batch size for high-density IoV. By leveraging Verification-Proxy Vehicles, it reduces verification delay by 90 percent, improves computation overhead by 91 percent, and increases verified messages by 67 percent, enabling scalable real-time IoV communication.

2. Intelligent Predictive Maintenance: Multivariate ML Model Optimization in an Edge-Fog-Cloud Environment
in Computing, 2025 [IF: 3.2]
Authors: Preeti Jain, Koustav Kumar Mondal, Debasis Das, Arpit Khandelwal
Link: https://link.springer.com/article/10.1007/s00607-025-01585-x

The paper presents FogBayes, an Edge–Fog–Cloud predictive maintenance framework using optimized multivariate ML models for low-latency automotive diagnostics. The system achieves 98.88 percent accuracy and an AUC of 0.9830, outperforming baseline solutions while ensuring scalable and efficient maintenance analytics.

3. A Blockchain-Integrated PUF Framework for Secure Authentication and Communication
in Ad Hoc Networks, 2025 [IF: 4.8]
Authors: Koustav Kumar Mondal, Debasis Das, Arpit Khandelwal
Link: https://www.sciencedirect.com/science/article/pii/S1570870525003075

This work integrates PUF-based hardware trust with blockchain-backed secret sharing and elliptic-curve cryptography to deliver a unified IoT security architecture. The framework ensures device-specific key generation, forward secrecy, and on-chain ephemerality, achieving sub-second authentication and scalable performance across diverse IoT deployments.

The official portal is accepting Applications for the TWO positions of 𝐉𝐮𝐧𝐢𝐨𝐫 𝐑𝐞𝐬𝐞𝐚𝐫𝐜𝐡 𝐅𝐞𝐥𝐥𝐨𝐰 at VANET Lab IIT Jodhpur.

** 𝐋𝐚𝐬𝐭 𝐝𝐚𝐭𝐞 𝐨𝐟 𝐀𝐩𝐩𝐥𝐢𝐜𝐚𝐭𝐢𝐨𝐧: 𝟏𝟎-𝟎𝟗-𝟐𝟎𝟐𝟓 **
👉 Apply directly from here: https://lnkd.in/g8J6QAkK

Revised age limit: 30 years

“𝐌𝐕𝐃𝐍𝐞𝐭: 𝐔𝐀𝐕-𝐛𝐚𝐬𝐞𝐝 𝐌𝐮𝐥𝐭𝐢-𝐦𝐨𝐝𝐚𝐥 𝐌𝐮𝐥𝐭𝐢-𝐕𝐞𝐡𝐢𝐜𝐥𝐞 𝐀𝐧𝐜𝐡𝐨𝐫-𝐅𝐫𝐞𝐞 𝐃𝐞𝐭𝐞𝐜𝐭𝐢𝐨𝐧” has been accepted and published as an Early Access article at the 𝘐𝘌𝘌𝘌 𝘛𝘳𝘢𝘯𝘴𝘢𝘤𝘵𝘪𝘰𝘯𝘴 𝘰𝘯 𝘝𝘦𝘩𝘪𝘤𝘶𝘭𝘢𝘳 𝘛𝘦𝘤𝘩𝘯𝘰𝘭𝘰𝘨𝘺 [IF: 6.1].

Heartiest congratulations to all the authors—

Nandini Saini, Department of Computer Science and Engineering, IIT Jodhpur
Devarsh Patel, Futuremachines
Dr. Debasis Das, Associate Professor, Department of Computer Science and Engineering, IIT Jodhpur
Dr. Chiranjoy Chattopadhyay, Associate Professor CSE, FLAME University

Publication link: https://lnkd.in/gG79GKue

The work presents a robust deep learning framework for detecting multiple vehicles in aerial images captured by UAVs, especially in noisy real-world environments. MVDNet introduces a novel Multi-Level Attention Module (MLAM) that enhances feature extraction by combining channel, spatial, and positional attention mechanisms, making the model highly resilient to sensor-induced noise such as Gaussian, speckle, and salt & pepper distortions. Unlike conventional anchor-based models, MVDNet adopts an anchor-free, one-stage detection approach that improves inference speed and localization accuracy. Extensive experiments on both single-modality (DOTA) and multi-modality (VEDAI) datasets, along with real-world UAV data, demonstrate that MVDNet outperforms state-of-the-art detectors, achieving 81.4% and 58.2% mean average precision respectively, with an impressive inference time of 14.4ms—highlighting its suitability for real-time, cloud-based or edge-deployable aerial surveillance systems.