moderation-all-resnext-2

Purpose

Image Moderation Classifier V2 model has been designed to moderate nudity, sexually-explicit, or otherwise harmful or abusive user-generated content (UGC) imagery. This is helpful to determine whether any given input image meets community and brand standards.

This model can identify 32 concepts in the following GARM Content Categories:

• Adult & Explicit Sexual Content
• Crime & Harmful acts to individuals and Society, Human Right Violations
• Death, Injury or Military Conflict
• Illegal Drugs / Tobacco / e-cigarettes / Vaping / Alcohol

Architecture

We based our image moderation classifier model on a ResNeXt architecture, and we trained it using a proprietary dataset gathered by our data strategy team. Given that ResNeXt is based on the residual ResNet approach, we have provided the following overview of both architectures:

ResNeXt

ResNeXt is a variation of the ResNet architecture that improves performance by increasing the width of the network. This is achieved by introducing a new dimension, cardinality C, which represents the size of the set of transformations. This approach is similar to the inception modules used in other architectures, allowing for more efficient processing and better performance.

ResNet

ResNet is a family of five ML CNN architectures that range from 18 to 152 layers, featuring one max pool and one average pool layer in each architecture, along with varying numbers of convolution layers.

Limitations

Due to the nature of our dataset, which heavily relies on real life imagery, the image moderation classifier model demonstrates weaker performance on hentai/cartoon harmful content.

Intended Use

Classification of inappropriate content which includes gore, drugs, explicit, suggestive, etc. in images and video. This model is intended to be used as a moderation tool to determine whether any given input image meets community and brand standards.

Resources

• ResNext on GitHub
• ResNext on Arxiv

Image Moderation Classifier V2 Author

Clarifai - GitHub