On-Device Content Moderation

TL;DR

This paper introduces a novel on-device NSFW content detection system that accurately identifies nude and semi-nude images, providing annotations and achieving high precision with minimal false positives, suitable for smartphone deployment.

Contribution

The paper presents a new on-device solution for NSFW image detection that includes semi-nude content moderation and unsafe body part annotations, with extensive testing on multiple datasets.

Findings

F1 score of 0.91 on custom dataset

0.92 MAP on NPDI dataset

0.002 false positive rate on safe image datasets

Abstract

With the advent of internet, not safe for work(NSFW) content moderation is a major problem today. Since,smartphones are now part of daily life of billions of people,it becomes even more important to have a solution which coulddetect and suggest user about potential NSFW content present ontheir phone. In this paper we present a novel on-device solutionfor detecting NSFW images. In addition to conventional porno-graphic content moderation, we have also included semi-nudecontent moderation as it is still NSFW in a large demography.We have curated a dataset comprising of three major categories,namely nude, semi-nude and safe images. We have created anensemble of object detector and classifier for filtering of nudeand semi-nude contents. The solution provides unsafe body partannotations along with identification of semi-nude images. Weextensively tested our proposed solution on several public datasetand also on our custom dataset. The model achieves F1 scoreof 0.91 with 95% precision and 88% recall on our customNSFW16k dataset and 0.92 MAP on NPDI dataset. Moreover itachieves average 0.002 false positive rate on a collection of safeimage open datasets.