Indu Cherukuri

Scene reconstruction and novel view synthesis:

• Performed experiments on improving reconstruction results from NeRF via multi sensor fusion methods.
• Performed research into 3D vision study topics and variations of NeRF to increase knowledge and provide valuable contributions.
• Collaborated with other researchers in formulating new problem statement with NeRF as base architecture.
• Implemented newly formulated problem and benchmarked its results against existing NeRF pipelines.
• Assisted in data capture for new benchmark data
• Processed and validated trajectory of collected data.
• Implemented and seamlessly integrated additional features to existing NeRF pipelines.

Document understanding: Extraction and mapping of information with respective field label in such a way that it can be mapped in a key value fashion.

• Responsible for developing and deploying to production multiple document understanding modules for following use cases
• Invoice Information extraction service
• Retail product Information extraction service, Datasheets Information extraction service
• Achieved 96% reduction in time required to process document/invoice from 45min to 1.5min
• 2 patents in proceeding to be filed
• Experimented with various research ideas using transformers, GNN architectures
• Lead development of synthetics module to generate realistic synthetic invoice images using TRDG library
• Experimented with various table detection techniques, image binarization techniques for PDF/images/invoices

• Automatic Fashion Cataloging: Developed end-to-end pipeline for a fashion service capable of classifying apparel and extracting all its attributes such as style, sleeve length, design, neck type with/for Indian apparels as main goal
• Trained models for detection and classification using SSD, ResNet.
• Achieved significant drop in time required to update product details
• Developed data collection tool using scraping techniques
• Worked on creating synthetics module to produce realistic looking apparels images with varied patterns using Graph Cut Segmentation, apparel detection model

• Developed software to automatically evaluate scripts and assign scores based on how close is code to correct solutions

• Developed Android application for food management which helps in reducing food wastage