Computer Vision and Deep Learning for anomalous object detection.
Endlessforms has done work with startups and enterprise level clients to detect anomalous objects in real time.
Client requested an automated way to select P & S wave arrivals from microseismic data. Endlessforms developed a Deep learning image segmentation model to select arrival crests, saving hundreds of hours picking arrivals by specialized geophysicists.
Client requested an object detection model that can identify and classify hair disease types from images.
An object detector was made in Pytorch and hosted on AWS App Runner via FastAPI and Docker for rapid prototyping in a production environment.
Client requested defect detector capable of micron-level resolution. This required pixel-by-pixel labeling of defects and clean pipe. A Pytorch script is Dockerized and invoked from a C#.NET application for fast ingestion and prediction.
Defect detector can identify galling on different pipe species and pipe thread (light blue) as well as new unidentified objects (green-yellow)
Model ensembling is used to verify accuracy of various models. Models converge on scratch in the example here.