International Conference of Image Processing (ICIP), 2023

A The study introduces a continual learning approach for pedestrian detection that can adapt to shifts in data distribution, addressing a common issue in current models. By modifying Elastic Weight Consolidation in an object detection network, the method maintains performance across different datasets, preventing catastrophic forgetting demonstrating significant improvements in miss rate reduction on the CrowdHuman and CityPersons datasets when compared to standard fine-tuning.

A multi-scale crowd counting and localization platform is proposed in this work. This novel architecture alongside with the multi-scale multi-task loss function have demonstrated promising performance.

We proposed the first Multi-scale gradient capsule GAN and utlized it for face super-resolution. This model outperformed state-of-the-art face super-resolution models.

In this work, the largest publicly available dataset for COVID19 is collected and a powerful COVID19 detection model based on CheXNet is proposed. This model can detect COVID19 accurately using meaningful features

One of the first attempts for high-scale super-resolution (8x) in biomedical domain. MSG-CapsGAN shows promising results in the medical domain as well.

A fully automated tooth extcartion model is implemented using genetic algorithm. A multi-feature extraction model with capsule classifier is developed for caries detection.

An efficient and scalable transistor level implementation of two different recurrent neural networks is proposed using memristor crossbar array.

Using LSTM network for user activity prediction in 5G network. The proposed model is capable of anomaly prediction one hour ahead.

The properties of the eroded area in the pipe is extracted from the pressure responce using GRU network. This work is the first erison dtection system using deep learning.