teenbta.blogg.se

Averaging the accuracies to get a better glimpse at the results.

Finding class-wise accuracy fog each category.Training our deep neural network model on the data.Dividing the data into training, validation, and test set.And obviously, we will deep neural networks in this article. Therefore, in this article, we will see how an imbalanced vision dataset affects class-wise accuracy. This will help you to have a better grasp of what we are doing today.Īlthough we got 95% accuracy on the test set, we did not check the class-wise accuracy. I would suggest that you read the previous article before moving on with this article. But I only showed how to get the final accuracy on the test set. I chose the problem as Caltech101 is a highly imbalanced vision dataset. In the last article, I laid out a detailed approach on how to get above 95% accuracy on the Caltech101 dataset. This is the result of the number of images in each class when the dataset is imbalanced. In image recognition, a deep neural network may predict 90% of one class correctly and only 20% of another class correctly. Deep learning algorithms suffer when the dataset is highly imbalanced. In this article, we will discuss how to get per-class accuracy in a highly imbalanced image/vision dataset.