I have the imagenet train, validation and test set. I have been able to assign each image in the validation set in its respective class folders with the help of some online resources. However, for the test set, I could not find the labels.
All the model evaluation experiments on Imagenet dataset I’ve seen people do only use a fraction of the validation set. I have not been able to find one that used the test set or a fraction of it.
I really want to use the imagenet test set to evaluate my model. How do I go about it or where can I get the labels for the test set?
I haven’t worked with ImageNet but a possible approach (in case you aren’t able to access test set labels) could be to partition your train set into training and validation divisions, and use the validation set as the test set.
I have actually done this by taking some samples from the validation set and making evaluation on it.
I am doing evaluation on ViT. It is mentioned in the paper that imagenet training set was part of the data used for the training. Though the usage of the validation set wasn’t mentioned but I want to believe that It might have been used for hyperparameter tuning.
So, I just want to make evaluation on samples I am very sure the model has not seen before.