I’m experiencing a strange error while reading data via ImageClassifierData.from_paths.
Here is how I tell it where to find train, validation, and test data:
It accepts train and validation dirs without any fuss, but when it comes to test data, it complains that the folder is empty or nonexistent:
But it exists, is not empty, and has the same 2-category structure as train and validation folders:
Any suggestion? It’s driving me crazy.
the test folder shouldn’t have same structure as the train and validation folders. in test folder there should be unlabeled pictures to classify. try putting your test jpegs into test folder without any subfolders in it.
Indeed: it is the whole point of test data . But I downloaded the dataset from kaggle, and the test folder, quite strangely, had that strange labeled structure by default.
Thank you anyway. I just merged the two folders, but the library should point out something like “…it is empty, does not exist, or does have an incorrect subfolder structure”.
indeed strange, when i download test images from kaggle they are without labels and in test1 folder without sub folders. as regarding your suggestion you could use github to report that. i like your suggestion.
