2

So I wanted to create a model that can classify a fruit into either "fresh" or "rotten". Currently I used this dataset from kaggle - https://www.kaggle.com/datasets/sriramr/fruits-fresh-and-rotten-for-classification

I then used LabelImg to apply bounding boxes for around 3,400 images (which consisted of fresh apples, rotten apples, fresh oranges, rotten oranges, fresh bananas and rotten bananas) but instead of using 6 different classes/labels, I used 2 labels called "fresh" and "rotten".

I got really good results for now but I want to scale up the model to be able to identify more fruits such as kiwis.

What would be the ideal way to do this?

I have tried adding images of fresh kiwis and rotten kiwis into the existing dataset and training a new model but the model still classifies fresh kiwi as rotten (im guessing its because browness is associated with rotteness for the 3000+ images from apples, oranges and bananas).

Do I make a new model with 8 labels? So "fresh apple, rotten apple, fresh kiwi, rotten kiwi, ...".

I am not sure how the bias comes into play when there are more images for one label than the others.

newprogrammer2000
  • 21
  • 1

1 Answers1

0

It would be better to increase the accuracy if you classify them with different labels. fresh apple, rotten apple, fresh kiwi, rotten kiwi etc. Thereafter you can cluster and print them in your code as fresh and rotten.

Makhele Sabata
  • 582
  • 6
  • 16
Fatih G.
  • 5
  • 2