Why the accuracy of pretrained resnet50 model is very low

Test: [2570/5000] Time 0.024 (0.060) Loss 19.2762 (18.3547) Acc@1 0.000 (0.117) Acc@5 0.000 (0.381)

This is the recognition result on test dataset of imagenet.

Start by reproducing the reported ImageNet evaluation with the pretrained ResNet-50 model and inspect the test setup, since the issue provides only a loss and accuracy log. Done means identifying why the reported accuracy is low and documenting or correcting the cause so the model reaches the expected evaluation result.

Written by the indexing model from the issue text.