Hi, I had a question regarding the use of pretrained ImageNet model for training the classifier.
The dataset website warns about using pretrained ImageNet models due to the possible overlap between the CUB testing images and ImageNet training images.
Is this potential overlap discounted for reporting the final results?
Hi, I had a question regarding the use of pretrained ImageNet model for training the classifier.
The dataset website warns about using pretrained ImageNet models due to the possible overlap between the CUB testing images and ImageNet training images.
Is this potential overlap discounted for reporting the final results?