Paper Image

Context-aware image mixing for industrial inspection

Published on:

18 January 2024

Primary Category:

Computer Vision and Pattern Recognition

Paper Authors:

Hyungmin Kim,

Donghun Kim,

Pyunghwan Ahn,

Sungho Suh,

Hansang Cho,

Junmo Kim

Bullets

Key Details

ContextMix mixes resized full images into other images to create new training data

It is designed to address data imbalance issues in industrial applications

The approach learns useful features from occlusions and image resizing

ContextMix improves performance across tasks with little computation cost

It demonstrates particular efficacy on a real industrial defect dataset

AI generated summary

Context-aware image mixing for industrial inspection

This paper introduces ContextMix, a data augmentation method tailored for industrial applications that mixes entire resized images into other images to generate new training data. This helps mitigate overfitting and performance issues stemming from the severe data imbalance common in manufacturing environments. ContextMix enables learning useful features from varying object sizes and occluded images with minimal computation cost. Evaluated on classification, detection and segmentation tasks, ContextMix demonstrates improved performance and robustness over existing techniques. Notably, it shows strong real-world efficacy on an industrial passive component defect dataset.

Answers from this paper

Comments

No comments yet, be the first to start the conversation...

Sign up to comment on this paper

Sign Up