Weights loaded through torch.hub.load give 0% performance.

I download imagenet data through Kaggle and loaded the swav pretrained weights using the torch.hub.load method.

After that, for evaluation, we tried to evaluate the imagenet val set as a swav model. But the performance is 10% Is it possible to receive data from Kaggle?(ImageNet Object Localization Challenge | Kaggle)

If anyone knows please help me.

If I understand correctly, the expected ~75% accuracy of the swav model is on the ImageNet 2012 validation set. However, it looks like the linked data is for a newer, different ImageNet competition.

10% accuracy seems to suggest that the data is from a totally different distribution and that the object categories are different. However, 10% seems low if the object categories are the same. Could you check the directory organization of the validation set and see if the class labels match with the classes in the 2012 dataset? You might be able to compare the directory names to check this (e.g., if the class directory names match those in this shell script: https://raw.githubusercontent.com/soumith/imagenetloader.torch/master/valprep.sh).