Syed Waleed Hyder

I am a research engineer at Retrocausal, Redmond, where I work on computer vision and machine learning.

At Retrocausal I work on Video Understanding. Particularly my research is on supervised and unsupervised fine-grained human activity segmentation at video and dataset level

Email  /  CV  /  Twitter  /  Github  /  Bachelor's Thesis

I'm interested in computer vision, machine learning, and image processing.

My research is on supervised and unsupervised fine-grained human activity segmentation at video and dataset levels using both frames and skeletons (pose information).

Slum mapping in satellite imagery using deep learning. Collected the image data on slums for Karachi and Islamabad, filtered the data. Used the Fully Convolutional Networks (FCN) to segment the slums from non-slums. Using loss functions to characterize the class imbalance since in majority of slum datasets the foreground pixels are dominant. An extension of this work is my bachelor's thesis (final year project)

SportsEye. AI-powered football analytics platform. Track every movement on the pitch using broadcast streams to run our analysis, without any extra hardware to be installed on stadiums. Player and Ball Detection using Single Stage Detectors. Player tracking and re-identification using deep sort. Broadcast view classification using custom CNNs. Field Line Masking using Pix2Pix GANs. Camera Pose estimation with classical Homography estimation. Timer localisation using EAST detector and recognition using OCR based methods.

RetailWiz. Targeted Ad platform for retail stores. Designed and implemented the AI pipeline for smart gondola capable of person tracking and profiling gender, age and emotions from the facial crop and use it for targeted advertisement. Optimized the pipeline for memory on Jetson Nano by writing a custom Python wrapper for TensortRT C++ (2 YOLOv5 detectors, 2 binary classifiers).

Adlytic. Audience Analytics platform for retail stores. Implemented the pipeline capable of footfall counting with age and gender classification, area-wise heatmap generation and dwell time. Developed APIs to help KPI dashboard consume the analytics. This system is deployed using docker in 50+ different retail stores and analytics are above 90% accurate. (YOLOv5, DeepSORT, BYTETrack).

Regeneron Pharmaceuticals. Platform for data analysis of Patient Wearable Devices. Developed a platform using Apache Nifi for data ingestion, PySpark for ETL and post-ingestion, Hive and AWS Redshift for analytics, and AWS S3 for cloud storage. Performed gait analysis using machine learning (ML) algorithms on time series-data from sensors of Moticon device. Classified the human activity using ML algorithms on time-series data from accelerometer of Actigraph device

This website design is taken from https://jonbarron.info/