Looking for Pre-trained Model for Image Categorization (Screenshots, Photos, Scans, etc.)

I’m on the lookout for a pre-trained model that’s capable of classifying images into various categories commonly found on a computer. The categories I’m interested in include, but are not limited to:

  • Screenshots
  • Photos
  • Scanned Documents
  • Graphics and Designs
  • Video Thumbnails

My goal is to find a model that understands and categorizes these images based on their nature and content. If a comprehensive model doesn’t exist, I’m also open to models that address subsets of these categories or any advice on how to approach training a model for this purpose.

Here are some specific questions I have:

  • Does anyone know of any pre-trained models that fit expectations?
  • Are there datasets available that could help in training a model for these categories?
  • Any recommendations on model architecture or training strategies for this kind of task?

Hi Vincent, I was wondering if you were able to find a model that suits your needs. I have similar needs for the work I am trying to do, but my research just started. If I find anything, I will make sure to post it here too. Thanks!

ResNet, Inception, VGG and MobileNet all have good results in classifying a wide range of images. To train a model for these categories, you will need datasets such as ImageNet, COCO, or OpenImages that contain a variety of images from different categories. In terms of model architecture and training strategies, it is best to start with a simple model such as MobileNet and gradually increase complexity using deeper architectures as needed.