Abishi Chowdhury

Abishi Chowdhury

Auto-generated profile, based on publicly available information

Abishi Chowdhury

Vellore Institute of Technology

Research Focus & Metrics

Academic Career Length

12Years (2015-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
28 May 2022 in Computers & Industrial Engineering
Verified
Journal Article
Closed access
ACAbishi Chowdhury
Abishi Chowdhury
SRShital Raut
Shital Raut
APAmrit Pal
Amrit Pal

14 August 2024 in IEEE Access
Verified
Journal Article
Open access
GAGovind Agarwal
Govind Agarwal
KJKritika Jindal
Kritika Jindal
ACAbishi Chowdhury
Abishi Chowdhury
VSVishal Krishna Singh
Vishal Krishna Singh
APAmrit Pal
Amrit Pal
In the rapidly evolving world of apparel, writing clear and interesting product descriptions is crucial to attract customers. In light of the importance of automated descriptions for apparel, this work explores the field of image captioning for apparel photos and expands its use to include captioning videos to enable visually impaired people to access and understand dynamic apparel content. To address the issue of diversity in datasets, we curated a collection of images that were divided into 26 classifications. With the use of Convolutional Neural Network (CNN) architectures like ConvNeXtLarge and Long Short-Term Memory (LSTM) architectures, our suggested system can automatically provide accurate and captivating captions for both still photos and moving videos that feature clothing. The LSTM network smoothly blends the visual data extracted by the CNN component from clothing photos and videos to produce captions that are both semantically and linguistically accurate. In addition, a YOLO model is included for real-time object detection, which makes it possible for the model to precisely identify and track several articles of clothing at once. The suggested architecture is evaluated using the BLEU score performance metric; research on the selected dataset yielded a BLEU-1 score of 0.983 for the ConvNeXtLarge-based model.

19 August 2025 in IEEE Access
Verified
Journal Article
Open access
PSPratyush Kumar Singh
Pratyush Kumar Singh
ACAbishi Chowdhury
Abishi Chowdhury
APAmrit Pal
Amrit Pal
One of the major challenges video surveillance systems usually face is high communication costs and computational bottlenecks, leading to latency in the transfer of high-definition video data. Such issues generally arise when large-scale projects are implemented in real-time capabilities. The paper explores Federated Learning (FL) algorithms for an efficient and scalable suspicious activity detection system to address these limitations. Federated Learning allows a decentralized approach that allows model training on participating devices (clients), and only the model updates are shared with the central server instead of sharing a huge amount of data. The approach reduces bandwidth consumption and minimizes latency without compromising the performance of the model. This study is among the first to explore federated learning (FL) for video-based anomaly detection using the UCF-Crime dataset. We propose and comparatively evaluate three FL aggregation strategies—FedAvg, Unweighted FedAvg, and FedSGD—under realistic, non-IID client data distributions. Unlike existing centralized methods, our approach offers enhanced privacy and reduced communication overhead while maintaining competitive performance. The results obtained in the experiments conducted on a varying number of clients for each approach suggest Unweighted FedAvg as the most reliable approach, which consistently achieves the highest accuracy and lowest loss for any given number of clients. The Accuracy for FedAvg ranges from 0.73 for 5 clients to 0.82 for 30 clients, and the loss decreases from 0.81 for 5 clients to 0.64 for 30 clients. Similarly, Unweighted FedAvg shows a consistent increase in accuracy from 0.74 for 5 clients to 0.85 for 25 clients and then drops slightly to 0.84 for 30 clients, and similarly for loss also it decreases from 0.78 for 5 clients to 0.53 for 25 clients and increases slightly to 0.56 for 30 clients. Despite the slight decrease in accuracy and increase in loss, it always performs better than FeDAvg. Whereas FedSGD has shown a very erratic pattern in change of accuracy and loss as the number of clients increases, with the lowest accuracy being 0.29 and the highest being 0.41, and for loss highest was 1.91 and the lowest was 1.59. The study successfully uncovers the potential of federated learning to address the limitations, such as computational bottlenecks, latency in data transfer, and model performance over non-iid data in real-time implementations of the public security systems.

9 September 2024 in IEEE Transactions on Consumer Electronics
Verified
Journal Article
Closed access
VYVasanth Yarlagadda
Vasanth Yarlagadda
ACAbishi Chowdhury
Abishi Chowdhury
APAmrit Pal
Amrit Pal
SMShruti Mishra
Shruti Mishra
SSSandeep Kumar Satapathy
Sandeep Kumar Satapathy
ADAshit Kumar Dutta
Ashit Kumar Dutta
Process mining, a burgeoning discipline within data science, demonstrates a significant contribution to the software development lifecycle of diverse real-time consumer-centric projects. This paper underscores the prominence of integrating predictive business process monitoring into organizational process models, as it can substantially impact profits and efficiency in any possible business domain along with improving services to consumers. The paper proposes a novel deep learning-based business process prediction model consisting of multiple layers with fine-tuning hyperparameters. The proposed model leverages input embeddings to represent each of the activities, and based on the training of the proposed model, the accuracy of the next activity is calculated. To assess the efficacy of the proposed model, it has been compared with the existing benchmark models. Our proposed model has shown a significant gain over the existing approaches. The results show that the proposed model outperforms these approaches by achieving an accuracy of 76% on the consumer helpdesk dataset along with an accuracy of 78% on the benchmark BPI dataset.

15 April 2025 in IEEE Access
Verified
Journal Article
Open access
RSRajan Singh
Rajan Singh
APAmrit Pal
Amrit Pal
SMShruti Mishra
Shruti Mishra
ACAbishi Chowdhury
Abishi Chowdhury
Though it is challenging, anomaly detection in real-time video data has significant implications in many disciplines like public safety, education, and agriculture. This paper proposes a novel approach to handle the complexity of dynamic real-world anomaly detection scenarios using three state of the art machine learning models: Convolutional Neural Networks (CNN), Region-based Convolutional Neural Network (R-CNN), and You Only Look Once (YOLO). The proposed system is precisely crafted to fast inform authorities of irregularities, so improving situational awareness and security procedures. Using trained models, applying the most effective one for real-time detection, and extensively investigating video data are part of the study method suggested here. The key objectives are to guarantee flexibility to match numerous circumstances and reduce false positives. First findings show the system’s good performance in numerous settings, thereby offering the basis for additional development and study. Especially, following rigorous training and testing, every model displayed varying degrees of accuracy. Emphasizing its effectiveness in focused detection tasks, the CNN model shown a remarkable accuracy of 73% in spotting anomalies inside specific areas of interest. R-CNN, on the other hand, shown greater performance in demanding environments and an accuracy of 80%. Especially appropriate for quick anomaly identification, YOLO, renowned for its speed and accuracy in real-time object detection, shone out with an amazing 91%. This work greatly helps the field of surveillance technologies by providing specific techniques to increase security and safety in useful surroundings. Emphasizing the transformational possibilities of novel anomaly detection systems, this work underlines the importance of constant research and deployment of innovative concepts to boost security and safety measures.

23 May 2025 in IEEE Transactions on Consumer Electronics
Verified
Journal Article
Closed access
VSVishal Krishna Singh
Vishal Krishna Singh
NANiharika Anand
Niharika Anand
APAmrit Pal
Amrit Pal
ACAbishi Chowdhury
Abishi Chowdhury
ASArjun Srivastava
Arjun Srivastava
Accurate anomaly detection in dynamic graph networks suffers due to lack of coverage of all aspects of information; specifically temporal, spatial and centrality based cross-coupled information. This work aims to address the challenge of precise and accurate anomaly detection in dynamic graph networks. It uses a graph-based diffusion technique to sample a fixed-size, yet cross-coupled, information-rich circumstantial node set for target edges. Centrality enabled spatial-temporal node encoding is considered as input to the dynamic graph based transformer network. The proposed method uses a set of four elements to make up the node encoding. The four encoding terms are combined to create an input that contains extensive centrality based cross-coupled spatial-temporal node encoding. The transformer module simultaneously captures all the required attributes with a single encoder. The performance of the proposed method is validated on six different datasets; UCI Messages, Bitcoin-Alpha, Digg Social, Enron Email, Epinions-Trust and AS-Topology. The proposed method outperforms the existing methods in terms of AUC-ROC score, accuracy, loss, and precision. Results show an improvement of 2.42% AUC-ROC value over the existing methods proving the models ability to counter over-fitting and provide accurate results.