Computer Vision Toolkit
What makes this innovative?
The core of this project is a state-of-the-art computer vision model based on the DETR model family developed by facebook. The novelty is its performance in low data regimes. This allows us to offer Computer Vision solutions for problems we otherwise needed far more data or time to tackle.
Data and Information Management
Image classification and object detection often require lots of training data and high manual effort
The toolkit achieves quick and presentable object recognition model with minimal effort
INITIAL SITUATION
Image classification and object detection are tasks that often require lots of training data and high manual effort to annotate them, while playing an important role in the digitalization and automation of processes in many industries.
SOLUTION
The computer vision toolkit is a modern framework based on a few-shot object detection transformer model, that is pre-trained on a large dataset and can be adapted to new use-cases with a relatively low amount of data.
The computer vision toolkit’s pipeline consists of an annotation tool, the training of the model and a simple interface to view the results of the inference. All of that is hosted on AWS and can be quickly adapted and deployed for our customers.
BUSINESS VALUE
Our few-shot object detection pipeline allows to quickly evaluate potential automation use-cases for object detection tasks at very low costs allowing our clients to explore new applications without high upfront costs.
FRAMEWORK & TOOLS
Python, AWS Sagemaker, Docker
Visit our YouTube Channel
