KaKao

As part of our advancement of the Kakao Map data and the development of a vision system that can be used in Kakao Map, we would like to share the development process of an image classification model that categorizes review images on Kakao Map.

 

The current image classification technology is considered a relatively simple task in deep learning. Various examples of implementation are easy to find. This article focuses on using processed pre-classified datasets with zero-shot classification to shorten the model development period and employing out-of-distribution (OOD) filtering to classify undefined categories. Thanks to the growth of deep learning models, frameworks, and the availability of various open-source models, there are now many strategies that can develop deep learning models more efficiently than just a few years ago. We hope this can be of some help to those adopting the deep learning model.

 

The entire article is available on the Kakao Tech Blog.