Can anyone please tell me how to download the complete imagenet dataset on which the pytorch torchvision models are trained on and their Top-1 error is reported on?
I have downloaded Tiny-Imagenet from Imagenet website and used pretrained resnet-101 model which provides only 18% Top-1 accuracy.
The torchvision classification models should have been trained on
ILSVRC2012, so you could download the dataset using the provided link.
Did you use the ImageNet example code from the PyTorch repository or how did you calculate the accuracy for Tiny-ImageNet?
@ptrblck thanks a lot for the reply. I first downloaded tiny-imagenet dataset which has 200 classes and each with 500 images from imagenet webpage then in code I get the resnet101 model from torchvision.models and perform inference on the train folder of tiny-imagenet. which provides only 18% accuracy as I mentioned earlier.
@ptrblck can u please tell me how to get ILSVRC2015 dataset seems like resnet101 pretrained models on torchvision trained on that.
I think it’s trained on ILSVRC2012 (sorry, had a typo earlier):
The pre-trained models are all based on Imagenet-12. Most of the model definitions also need 224x224 as the minimum input size. For CIFAR, you are better off with other repos.
The provided link should be a Download section.