Multiple facets of categorization

Part 1 (2017)
1

@jeremy This might be a dumb question but i am a bit confused so here goes.

Say I have a set of apparel images. Each of these can be categorized on multiple dimensions.

  1. Color (10s of color hues could be present)
  2. Top/Bottom (shorts, shirts, blouses, jeans, pants)
  3. Work/Casual
  4. Textile: silk, cotton

and so on…

Each image would have one or more of each type.
So how can the training data set be structured to be able to identify the values of all these dimensions for each image or the other way to look at it might be that each image might need to be present in multiple of these folders?

1 Like
2

That a most excellent question!

I am hoping to show an example of just such a model on Monday :slight_smile:

2 Likes
3

@jeremy would this be best done with the collaborative filtering method we learnt?

4

No - you should use the multiple-output model we saw in lesson 7. We used such a model to predict both bounding boxes and classes at the same time. You could use a similar approach to predict categories across all the dimensions you’re interested in.