Efficient Pipeline for Image Classification
Camera-based recognition systems are widely-spread in manufacturing, logistics, robotics as well as retail and the public sector. Image recognition systems enable measurement of object flows, surveillance and monitoring, quality control and inspection and also automatic check-in/out systems. Machine Learning and its advancements are the current state of the art for image classification on the software side.
INTRODUCTION
Problem: These current technologies, however, typically require massive amounts of training data and an extensive training process before being able to put into the field. This is an insufficient approach for industrial applications that rely on repeatedly changing product portfolios or lack of pre-existing information-sets.
Solution: EPIC enables the user to show new items to the system which learns to recognize them immediately.
Few-Shot Learning for Highly-Flexible Object Detection
Our EPIC software framework capitalizes few-shot learning technologies for industry-grade applications providing highly-flexible recognition systems for data-efficient as well as adaptable usage.
Appreciated amongst experts: EPIC won the Deep Tech Award 2019
A few simple steps and just a handful of images shot take an unknown object to successful identification in the system. EPIC recognizes the new object immediately after the registration.
Features
Let system learn new objects within seconds with just a handful of images
Implementation as embedded camera application as well as a scaling cloud service
Works with nearly every camera from most of all manufacturers
Best fit for applications with (frequently) alternating product range
Our human centric user interface guides the operator step-by-step through the easy process of creating new entries as well as editing and deleting existing ones.
APPLICATIONS & USE CASES
Machine learning based visual check-in/out systems are currently conquering inventory assessment as well as cash register systems. In comparison to classic technologies, there is no longer a need for scanning artificial markers, e.g. QR-codes or RFID tags. The training of novel objects could be implemented flexibly even from remote locations with hand-held devices. Consequently, novel products can be learned within seconds through cutting-edge few-shot learning in distributed computing scenarios.
Quality Assurance and Commissioning with EPIC: different items in the production line can be identified by robot-mounted camera and trigger specific actions to handle the corresponding item
Automatic Order Processing with EPIC: inventory of all kinds can be identified and processed automating systems like grocery store conveyor belts and cafeteria cast registers
Inventory and Flow of Goods with EPIC: monitoring the coming and going of inventory becomes efficient and straightforward tracking the flow at every threshold
Our Quote
The EPIC software framework is available as a software development kit (SKD) and web service for usage in different industries. In addition, we offer consulting services for few-shot learning technologies as well as tailored project-based solution development.
Drop us a message info@gestalt-robotics.com or give us a call at +49 30 616 515 60.