Image Quality Control for an
Standardizing the quality of content on an e-commerce marketplace is now essential to maintain a good experience for a customer. Marketplaces today can rack up thousands of dollars in expenses and hours of manual labor to quality control or moderate the image and text content being generated on their sites.
With the Nanonets API, companies can entirely automate various aspects of the QC or moderation processes using highly customized rules with minimal data and implementation effort.
We helped an e-commerce selling & re-selling platform score the quality of uploaded images and detect and filter out all images depicting explicit content.
We trained the following models with client images and customized them with specific rules and ensured a 95% accuracy rate.
Image Quality scorer
The model applied a percent score for image quality, taking into account resolution as well as overall quality - including angles, shadows, lighting, sharpness, etc. Images that scored below 40% were immediately rejected, ones that scored above 80% were immediately accepted, while those between 41% and 79% were marked for manual review.
NSFW moderation model
The model automatically rejected all images that contained any of the following type of content: hardcore pornography, softcore pornography, explicit nudity, suggestive nudity, violence, and gore.
Image Quality Model: 1000 high quality images and 1000 low quality images.
NSFW Model: 2000 images balanced between the various NSFW categories and 2000 acceptable images as a control group.
The model was deployed on the customer’s premise via a docker set up. The machine is capable of processing an image within 50 milliseconds. The customer wanted to ensure that the data never left their premises. Running the models on their own infrastructure allowed them to maintain data compliance.
Note: Nanonets also hosts its own models for even quicker implementation. With this option, we can host models on the Nanonets cloud and customers can access them through API calls made over the web.
- Improved user experience on the platform.
- Maintained sanity of content.
- Saved time and money by automating a manual and time consuming task.