Top 5 Pre Trained Models In Image Recognition

Image recognition refers to an up-and-coming computer technology that works to process a selected image and give information about the objects found in it. There are specific applications for this type of software throughout a host of industries, so let’s take a closer look at the pre-trained models that can be used.

Image recognition models

Models can be created to define the subset of image detection where it needs to be applied. The models currently on offer at are as follows:


The general classification model covers the overall content of a selected image and helps to give basic information to users who are unsure of what they’re looking at.

Not Safe For Work (NSFW)

Not Safe For Work (or NSFW) is the classification given to images that contain nudity and similar content that may need to be obscured or filtered out for younger audiences.


This pre-trained model can help to define locations within images for fast identification. As an additional feature, there is a predicted score percentage to give users an idea of how relevant the result is likely to be.

General Object Detection

This differs from the general model as not only does it detect the object in the selected image, but also its potential/exact location.

Pose estimation

When images contain people, there are times when their poses can identify key information. This model can calculate as many as 15 joints and key points of the human body for optimum results.

Optical Character Recognition (OCR)

This model focuses on finding text within an image, and can detect a host of languages to ensure nothing gets missed.

Top 5 use cases and the model they can benefit from

  1. Content Moderation uses NSFW, general classification and general object detection models
  2. Goods Detection and Classification uses general classification and general object detection models
  3. Automatic Hashtags uses places, general classification and general object detection models
  4. People Counting uses the general object detection model
  5. Augmented Reality/Game Development uses the pose estimation model

As this is just a basic run down, you can find more info here.