Publications

Multimodal Humor Detection in Videos via Domain-Specific Representations and Deep Hybrid Fusion Networks

Jamuna S. Murthy, G. M. Siddesh

Soft Computing (Springer Nature), 2026

Humor in videos is hard to detect because cues are spread across text, visuals, and audio that are misaligned in time, imbalanced, and noisy. This framework combines Domain-Specific Representation Learning, which separates modality-invariant and modality-specific features with a Bi-LSTM, and a Deep Hybrid Fusion network that merges them through cross-modal attention with suppression, reaching up to 95.8% accuracy on IEMOCAP, CMU-MOSI, and CMU-MOSEI.

CAMMEL-ViT: Context-Aware Memory with Lightweight Vision Transformers for Medical Image Segmentation

Abhay Bhandarkar, Jamuna S. Murthy (contributing author), Amogh Kalasapura

MIUA 2026 — Medical Image Understanding and Analysis, University College Dublin, 2026

CAMMEL-ViT is a dimension-agnostic CNN–Transformer architecture for 2D and 3D medical image segmentation. It pairs multi-scale convolutional encoding with an Efficient Lightweight Vision Transformer (EL-ViT) for reduced-resolution global context and a Context-Aware Memory Module (CAMM) that injects a compact global memory prior for anatomically coherent predictions, improving Dice and HD95 on BraTS, ACDC, and BTCV at favorable computational cost.

Video

KathaTrace: Diagnosing Semantic Trajectory Collapse in Generated Visual Narratives

Jamuna S. Murthy, Amin Karimi Monsefi, Rajiv Ramnath

Under submission to a CORE A* venue · arXiv preprint, 2026

Recent story generators produce visually coherent image sequences, but coherence does not guarantee that the transition meaning of the source story survives. KathaTrace is a generator-agnostic protocol for diagnosing semantic trajectory collapse, evaluating each transition under text-only, image-only and text-plus-image evidence. It contributes KathaBench-25K and defines the Semantic Trajectory Gap, the transition meaning lost during visualisation.

Paper Project page

Controlla: Learning Controllability via Graph-Constrained Latent Geometry

Jamuna S. Murthy, Amin Karimi Monsefi, Rajiv Ramnath

Under submission to a CORE A* venue · arXiv preprint, 2026

Controllable multimodal generation is usually posed as inference-time conditioning, which leaves unstructured how semantic attributes evolve and invites identity drift. Controlla instead treats controllability as a property of latent geometry: it learns identity and attribute factors from multimodal inputs and aligns them with graph priors through graph-constrained optimal transport, so attributes follow graph-consistent trajectories while reference identity is preserved. The paper also contributes AffectHuman-43K, a leakage-aware multimodal benchmark.

Paper Project page

Towards Secure Video Surveillance: A Few-Shot Spatiotemporal Perception Transformer for Unseen Behavioral Anomalies

Jamuna S. Murthy, Dhanashekar Kandaswamy, Wen-Cheng Lai

IEEE AVSS 2025 (CORE A) — International Conference on Advanced Visual and Signal-Based Systems, 2025

FewShot-SPT is a transformer-based framework for few-shot anomaly detection in surveillance. It introduces Event-Guided Keyframe Extraction, Adaptive Modality Gating for multimodal fusion, and adaptive prototypical learning for generalization to unseen anomalies, reaching 91.6% AUC on UCF-Crime and 84.2% accuracy on XD-Violence.

Paper PDF Code

Multimedia Video Analytics using Deep Hybrid Fusion Algorithm

Jamuna S. Murthy, Siddesh G. M.

Multimedia Tools and Applications (Springer), 2025

A Deep Hybrid Fusion (DHF) algorithm for multimodal video analytics that integrates text, image, and audio features, using BiLSTM-based modality representations and attention-based fusion to reach 95.84% accuracy on benchmark datasets for humor detection and multimodal understanding.

Paper PDF Code

A Smart Video Analytical Framework for Sarcasm Detection using Adaptive Fusion and SarcasNet-99

Jamuna S. Murthy, Siddesh G. M.

The Visual Computer (Springer), 2024

A multimodal sarcasm detection framework integrating text (Enhanced-BERT), image, and audio features. A novel adaptive fusion strategy combined with SarcasNet-99 achieves 97.3% accuracy on large-scale video datasets.

Paper PDF Code

ObjectDetect: A Real-Time Object Detection Framework for Advanced Driver Assistant Systems Using YOLOv5

Jamuna S. Murthy, Siddesh G. M., Wen-Cheng Lai, et al.

Wireless Communications and Mobile Computing (Wiley), 2022

A real-time object detection framework for Advanced Driver Assistance Systems that combines vision and LiDAR-based perception with YOLOv5, reaching about 95% accuracy while running markedly faster than YOLOv3 and YOLOv4.

Paper PDF Code

TwitSenti: A Real-Time Twitter Sentiment Analysis and Visualization Framework

Jamuna S. Murthy, Siddesh G. M., K. G. Srinivasa

Journal of Information & Knowledge Management, 2019

Real-time sentiment modeling over streaming social media, integrating NLP pipelines with scalable distributed systems (Apache Storm, Redis) for low-latency inference and interactive visualization.

Paper PDF Code

A Real-Time Twitter Trend Analysis and Visualization Framework

Jamuna S. Murthy, Siddesh G. M., K. G. Srinivasa

International Journal on Semantic Web and Information Systems (IJSWIS), 2019

A real-time Twitter trend analysis framework that combines streaming data processing with semantic analysis and interactive visualization to identify emerging trends and track their momentum.

Paper PDF Code