Request for training script of Resnet-50 (Vanilla) for fine-grained dataset CUB_200_2011, Aircraft, StandfordCar, Caltech

[roy-ch]

I am writing to kindly request access to the train.py script that you utilized in your experiments on the fine-grained dataset. I am particularly interested in reproducing the results presented in Table 3 for the ResNet-50 vanilla model from your study.

I have attempted to train the model using the repository available at https://github.com/cyizhuo/Fine-Grained-Image-Classification/tree/main. When setting resnet-50 pretrained to True, I achieved the following test accuracies:

Dataset	acc
CUB-200-2011	86.74
FGVC Aircraft	93.25
Stanford Cars	94.09

I also have trained with setting resnet-50 pretrained = False, but the test ACC is to be Dataset	acc
CUB-200-2011	59.1
FGVC Aircraft	on going
Stanford Cars	77.68

Unfortunately, none of these results align with the accuracies reported in your Table 3.

[3neutronstar]

I hope this following commit would help. I train the resnet-50 from the scratch.

link