— Leading the Computer vision team for model's selection, training, and finetuning,
— Searching/designing and preparing datasets for models training,
— Research, Design, development, and training of Computer Vision models for the following modules:
Identification of guns, pistols, knifes in video feeds,
People detection and recognition:
use case: face recognition - if the person was previously caught on scamming,
use case: more than 1 person present near the screen of ATM,
Behavior recognition models:
use case: person in the video frame is trying to fix/scam ATM device,
use case: fight detection,
— Participating in architectural design of a highly efficient modular platform for distributed processing of raw and compressed video feeds from 21k-200k ATMs worldwide.
— Participating in the process of testing of trained models in the fields and incremental quality improvement.
— Interaction with the product owner.
Mediapipe framework (Skeleton models and behavior recognition)
Experience with models on edge devices
2+ years in CV applications;
previous experience in building high load computer vision applications;
previous experience in training and deploying of light models on EDGE device;
— Prior knowledge of C/C++/C#/Go as second language;
— Experience of development and deploying large scale distributed computer vision applications;
— Experience of development and deploying large scale distributed projects with Machine and Deep learning models;
— Experience with PyTorch and/or Tensorflow or other ML/DL libraries, (DLIB, mediapipe, scipy);