CVNodes: A Visual Programming Paradigm for Developing Computer Vision Algorithms

Advances in machine learning have led to a rapid pace of innovation in Computer vision and deep learning classification algorithms. Deep learning classification models are often limited in flexibility due to their fixed pre-processing steps embedded into the algorithm and lack ways to easily iterate...

Full description

Saved in:
Bibliographic Details
Published in2020 17th Conference on Computer and Robot Vision (CRV) pp. 174 - 181
Main Authors Wang, JunFeng, Hogue, Andrew
Format Conference Proceeding
LanguageEnglish
Published IEEE 01.05.2020
Subjects
Codes
Computational modeling
Computer vision
Deep learning
Machine learning algorithms
Programming
Visualization
Online AccessGet full text

Cover

More Information
Summary:Advances in machine learning have led to a rapid pace of innovation in Computer vision and deep learning classification algorithms. Deep learning classification models are often limited in flexibility due to their fixed pre-processing steps embedded into the algorithm and lack ways to easily iterate, debug, and analyze developed algorithms without programming knowledge. The lack of high-level tools for developing vision algorithms leads to longer development times that require significant knowledge of underlying algorithms. What about individuals without this deep knowledge of machine learning and vision yet wish to develop algorithms to prototype ideas? What about nonprogrammers such as designers and artists that wish to utilize the state-of-the-art in computer vision in their work? To address this under-served community, we propose a visual-programming solution akin to those found in modern game engines geared towards computer vision algorithm development. These results in a new prototyping tool to empower researchers and nonprogrammers to easily iterate algorithm development, use pretrained classification models, and provide statistical post-analysis tools.
DOI:10.1109/CRV50864.2020.00031