Open lassloss opened 1 year ago
The model of the global part and the local part in your code seems to be the same, and the pre-trained resnet101 you used seems to be fine-tuned. After I downloaded the pre-trained resnet101 from pytorch, the model’s results dropped by nearly ten points.
Have you reproduced the approaching situation?
The model of the global part and the local part in your code seems to be the same, and the pre-trained resnet101 you used seems to be fine-tuned. After I downloaded the pre-trained resnet101 from pytorch, the model’s results dropped by nearly ten points.