Training accuracy, Validation accuracy, Test accuracy

I’m new to deep learning and image processing. Can someone explain me this table?


I’m very appreciated!

Hey @huythonghust1711 , Welcome to the community.
This seems to be a part of some questions, While I am not aware of the whole question, it seems to compare different accuracies of a model.
I am taking an example here, suppose you were given a set of images/pictures of Dogs and Cats and asked to make a model and classify the images into dogs and cats. Then the above terms would mean.

  • Human Accuracy: How accurately a human like you and I are able to classify the set of pictures into Dogs and Cats.
  • Training Accuracy: How the model is able to classify the two images during training on the training dataset.
  • Valid Accuracy: How the model is able to classify the images with the validation dataset. ( A validation dataset is a sample of data held back from training your model that is used to give an estimate of model skill while training the model)
  • Test Accuracy: How the model is able to classify between the cats and dogs when real-world data or test data used.

Hope this helps! Thank You.

What I can see here is a comparison table that demonstrates different model’s training, validation and test courses based on accuracy metric plus the human determinations. The evaluation on validation data and testing data achieved quite different results, they are pretty consistent though. People might choose the 4th one as the final model as it yielded the most promising accuracy on validation and testing data.