Himanshu Singh

Himanshu Singh

Auto-generated profile, based on publicly available information

Himanshu Singh

Indian Institute of Technology, Jammu, Jammu, India

Research Focus & Metrics

Academic Career Length

5Years (2022-Present)

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

Representative Publications

View All Publications
17 November 2022 in IEEE Transactions on Artificial Intelligence
Verified
Journal Article
Closed access
HSHimanshu Singh
Himanshu Singh
SSSaurabh Suman
Saurabh Suman
BSBadri Narayan Subudhi
Badri Narayan Subudhi
VJVinit Jakhetiya
Vinit Jakhetiya
AGAshish Ghosh
Ashish Ghosh
Several research works have been developed in the area of action recognition. Unfortunately, when these algorithms are applied to low-light or dark videos, their performances are highly affected and found to be very poor or fall rapidly. To address the issue of improving the performance of action recognition in dark or low-light videos; in this article, we have developed an efficient deep 3D CNN based action recognition model. The proposed algorithm follows two-stages for action recognition. In the first stage, the low-light videos are enhanced using Zero-Reference Deep Curve Estimation (Zero-DCE), followed by the min-max sampling algorithm. In the latter stage, we propose an action classification network to recognize the actions in the enhanced videos. In the proposed action classification network, we explored the capabilities of the $R(2+1)D$ for spatio-temporal feature extraction. The model's overall generalization performance depends on how well it can capture long-range temporal structure in videos, which is essential for action recognition. So we have used a Graph convolutional network (GCN) on the top of R(2+1)D as our video feature encoder which captures long-term temporal dependencies of the extracted features. Finally, a Bidirectional Encoder Representations from Transformers (BERT) is adhered to classify the actions from the 3D features extracted from the enhanced video scenes. The effectiveness of the proposed action recognition scheme is verified on ARID V1.0 and ARID V1.5 datasets. It is observed that the proposed algorithm is able to achieve $96.60\%$ and $99.88\%$ as Top-1 and Top-5 accuracy, respectively, on ARID V1.0 dataset. Similarly, on ARID V1.5, the proposed algorithm is able to achieve $86.93\%$ and $99.35\%$ as Top-1 and Top-5 accuracies, respectively. To corroborate our findings, we have compared the results obtained by the proposed scheme with those of fifteen state-of-the-art action recognition techniques.

17 December 2025
Verified
Conference Paper
Open access
PKPrabhat Kumar
Prabhat Kumar
HSHimanshu Singh
Himanshu Singh
BSBadri Narayan Subudhi
Badri Narayan Subudhi
GCGaetano Di Caterina
Gaetano Di Caterina
VJVinit Jakhetiya
Vinit Jakhetiya
TVT. Veerakumar
T. Veerakumar
Spiking Neural Networks (SNNs) represent a promising approach for energy-efficient, event-driven computation inspired by biological neural processes. However, their practical deployment is often hindered by three major challenges: limited availability of event-based datasets, poor interpretability of model decisions, and lack of reliable mechanisms for estimating prediction uncertainty. These limitations restrict the usability of SNNs in real-world, safety-critical applications such as robotics and edge-AI. To address these issues, this work presents a novel framework that integrates two complementary techniques: event-based data augmentation (EDA) for enhancing data diversity in neuromorphic datasets and evidential deep learning (EDL) for modeling uncertainty using a Dirichlet distribution over class evidence. We utilize a VGG-11-derived architecture, adapted into a spiking format using leaky integrate-and-fire (LIF) neurons. This architecture is trained end-to-end using surrogate gradient techniques across fixed timesteps, with temporal normalization employed to stabilize spike-based learning. An EDL module is appended to the output layer to provide calibrated confidence estimates alongside class predictions, this module is optimized using a combination of cross-entropy and KL-divergence-based evidential loss functions. We have evaluated the performance of our proposed system on two benchmark neuromorphic datasets, namely CIFAR10 DVS and N-Caltech101. Experimental results show that our proposed framework achieves classification accuracies of 76.57% and 77.20% on CIFAR10-DVS and N-Caltech101 datasets, respectively. Furthermore, we compare our technique with five state-of-the-art techniques on the aforementioned datasets, respectively, confirming the superiority of the proposed framework.