CODEWITHSUNDEEP
pythonpytorchconv-neural-networkcomputer-science
Daniel
Daniel

Reputation: 1

CIFAR-10 CNN using PyTorch

I'm working on CIFAR-10 dataset using PyTorch and observed some categories have low accuracy around 40-50%. Why do some categories of image were more difficult to work with?

enter image description here

Upvotes: 0

Views: 250

Answers (1)

Snehal Patel
Snehal Patel

Reputation: 192

The reason for low accuracy for some classes often has to do with significant similarity with other classes. For example, cat and dog may be confused for each other fairly often by the model. Your confusion matrix shows what types of mistakes your model is making. To some extent, it is confirming this intuition that cats and dogs are often confused (as wells as for other animals). I would suspect that the similarity in the "domestic" background of these domesticated animals is also contributing to the confusion.

Upvotes: 1

Related Questions