Closed yaolu-zjut closed 1 year ago
Ah yes, it is different on their repo.
We used the same train and validation splits as defined by the original ImageNet dataset.
FastAI originally did it this way as well, but at some point they released a new version of the splits.
See here: https://github.com/fastai/imagenette/blob/master/README.md?plain=1#L13
Sorry for any confusion, and let me know if I can be of any more help!
Hello, I am very interested in MTT. When I use your code about imagenette, I find it is different from the original imagenette, am I right?
See imagenette in https://github.com/fastai/imagenette