sindhurakshit / glide-text2image-inpainting Goto Github PK

A dedicated technology leader driven by a hunger for innovation and the development of new products and technologies. With a academic background in Computer Science and Artificial Intelligence, coupled with senior management experience, he is passionate about pushing the boundaries of technology and creating impactful solutions. Kumar is eager to contribute to dynamic environments where innovation thrives and where he can continue to learn and grow while making a positive difference.

• Python (Pytorch, Numpy, Scikit, Matplotlib, TensorFlow, Pandas, OpenCV, Flask)
• Additional Libs (tqdm, librosa, torchvision, torchaudio, os, sys, re, pickle, glob, shutil, urllib, zipfile.)
• Computer Vision, UNet, Cell Segmentation, Object Segmentation.
• Deep Learning, Transformers, Generative Adversarial Networks (GANs).
• Recurrent Neural Networks, Reinforcement Learning, Search Graphs.
• GitHub, SVN, Jenkins. •

• Full-stack developer.
• Java (Spring Boot, Hibernate, JPA, Oauth2)
• Restful API, JavaScript/JSP , Thyme leaf.
• PL/SQL, MySQL, Oracle, MariaDB, Mango DB, Redis.
• XML/JSON, C/C++.
• Cloud Computing (AWS), Kubernetes, MLFlow.
• Robotics – Motion Planning, Robotic Vision, Visual Servoing.
• RoS (python, catkin, moveIt, rviz, gazebo, services, messages, topics, rqt, package creation, publishing).

• Technical Product Management.
• Project Management
• Agile/Scrum Software Development.
• Customer Management.
• Software Product Development.
• Team Recruitment and Development.

Immune Cell Segmentation
Cell segmentation is the important task of identifying different cells in a microscopic image, this is precursor to measuring important parameters like ratio of different cell types, size of cells, nucleus size, nucleus to cell size ratio which help predicting various diseases and stages. Whole-slide images bring specific challenges in processing due to hierarchy and huge size sometimes a single file could be more than giga-bytes.

In this project developed a machine learning pipeline to split large whole-slide images for segmentation using state-of-the-art transformer based deep learning algorithms (Swin-Unet) for pancreatic cell segmentation and nuclei counting using Pytorch, Pandas, Openslide, NumPy, OpenCV as core libraries.
Git: https://github.com/sindhurakshit/cell-segmentation.

Facial Expression Recognition (FER)
Facial Expression Recognition (FER) is a technology endeavour to detect a person’s expressions by analysing static image or video with potential use cases in education, public- safety, crime detection, market research , personalised cars, engaging video games and many more. Designed a deep learning model based on VGG19 network with additional three fully connected layers and SoftMax for classification. We applied data augmentation techniques like random horizontal flipping, random vertical flipping, colour jittering , random erasing and centre cropping and further optimised model with ablation study. Key libraries used in the project are torch, torch vision, NumPy, pandas, matplotlib.
Git: https://github.com/sindhurakshit/FER.

Image Synthesis and Inpainting with Generative Models (GLIDE &CLIP)
One of the way people reflect creativity is by visual representations, such as illustrations, paintings, and photographs. These artefacts can often be easily described in natural language, but generally require specialized skills and hours of labour to create. Researched and analysed various approaches of text to image generation including Generative Adversarial Network (GAN), Variational Encoder (VAE) and diffusion based models. The open AI GLIDE and CLIP was chosen because enhanced technical capabilities including photo realism. Implemented the photorealistic text to image generation and inpainting systems using diffusion based generative models, using pre-trained, Open AI GLIDE (Guided Language to Image Diffusion for Generation and Editing) and CLIP (Contrastive Language–Image Pre-training) models.
Git: https://github.com/sindhurakshit/GLIDE-Text2Image-inpainting.

Advanced Robotics with RoS Feb 2022 to May 2022 This project was implemented with objective of learning modular RoS programming and advanced robotic concepts. A custom catkin package consisting, different message types, services, topics for calculating, publishing and plotting positional, velocity and acceleration of cubic polynomial trajectories using rqt plot and rosrun/roslaunch was developed.
Git: https://github.com/sindhurakshit/advanced-robotics-with-ros.

MLEnd Hums and Whistles
The objective of this project was identifying songs based on humming and whistles, an end-to-end machine learning project with involvement in conceptualisation, data-set creation, data cleaning and implementation of feature extraction with and audio classification using support vector machines (SVM). The key machine learning pipeline stages involved loading audio files, resizing each file to fixed size, adding audio augmentation methods, feature extraction and SVM classifier.
Git: https://github.com/sindhurakshit/MLEnd-Hums-and-Whistles

Data protection & Privacy Safeguards
Current digital age is witnessing an exponential proliferation of sophisticated hardware- and software-based intelligent solutions that are able to interact with the users at almost every sensitive aspect, collecting and analysing a range of data individual. This data, or the derived information are often too personal to fall into unwanted hands, and thus users are almost always wary of the privacy of such private data that are being continuously collected through these digital mediums. This study aimed to analyse the UK GDPR framework (Legislation.gov.uk, 2018) and other frameworks with focus on data protection and privacy by design identify the key challenges and proposed solutions. This research is based on a systematic literature review specifically in telecom, heath care, financial domains. The deliverables of this research were a video presentation and a research paper.

Paper: https://github.com/sindhurakshit/EAD/blob/main/DATA%20PROTECTION%20%26%20PRIVACY%20SAFEGUARDS%20-%20Kumar%20Sindhurakshit.pdf
Video: https://youtu.be/hHLWjkeN4vg/.

Fetal Brain Segmentation
The primary objective of this project is to develop a robust and accurate model capable of segmenting fetal brains from medical imaging data. Achieving accurate segmentation is crucial in various medical applications, including prenatal diagnosis, fetal health monitoring, and understanding developmental abnormalities. To accomplish this objective, the model is developed utilizing the HC18 dataset, a comprehensive collection of fetal brain images specifically curated for research and analysis purposes. Git: https://github.com/sindhurakshit/FetalBrain

Coming soon...

• Multi-tenant counterfeit product detection. IN 422176