Arkaprabha Basu

Solving edge solutions for video surveillance. Employed YOLOv8 and YOLOv5 in PyTorch for object detection, segmentation, and classification. Utilized Tao ETLT and ONNX for efficient model preparation, enabling the isolation of anomalous objects. Applied SlowFast Network for recognizing anomalous activities and Re-Identification for person tracking across cameras. Deployed via TensorRT conversion for Nvidia-Deepstream.

1. Temple Tile Reconstruction using Multimodal analysis (Heritage Restoration: Bankura Terracotta Tiles). Revitalized using a 3D geometry-aware GAN for improved reconstruction with context-aware few-shot generation and Super Resolution. Automated detection of defective tiles and replacement enhances efficiency. The solution is deployed through a python Streamlit application. 2. Explanation of Lung Diseases using machine learning tools (A robust solution for COVID-19 Identification from Chest X-rays). Enhanced CXR analysis using Image enhancement, ResNet50-based CNNs, and a class imbalance algorithm for improved detection. Employed Explainable AI methods for detailed analysis, intensifying the region of abnormality in the CXR rib cage.