Bootstrapping Deep Neural Networks from Approximate Image Processing Pipelines

TL;DR

This paper presents a method to automatically generate large labeled datasets from existing image processing pipelines, enabling effective training of deep neural networks that can outperform or match traditional components with less manual labeling.

Contribution

The authors introduce a workflow that leverages existing pipelines to create labeled data automatically, facilitating deep neural network training without extensive domain knowledge.

Findings

Deep neural networks trained with automated labels achieve comparable or better performance.

The approach reduces the need for manual labeling and can lower computational costs.

Effective even with small initial labeled datasets.

Abstract

Complex image processing and computer vision systems often consist of a processing pipeline of functional modules. We intend to replace parts or all of a target pipeline with deep neural networks to achieve benefits such as increased accuracy or reduced computational requirement. To acquire a large amount of labeled data necessary to train the deep neural network, we propose a workflow that leverages the target pipeline to create a significantly larger labeled training set automatically, without prior domain knowledge of the target pipeline. We show experimentally that despite the noise introduced by automated labeling and only using a very small initially labeled data set, the trained deep neural networks can achieve similar or even better performance than the components they replace, while in some cases also reducing computational requirements.