cswluo / crossx Goto Github PK

Hi,

Thank you for the code!

I wanted to know if you have with you hyperparameters to train a model from scratch on cub-200-2011 or vgg-aircraft? I tried using the default values however, they seem to be giving very bad performance.

Thanks.

As the title, both of pretrained models are invaliding, can you fix it? please.

Hi Luo!

Thanks for open sourcing the code in PyTorch! I was wondering what the license of the repo is?

Cheers

Hi @cswluo , thank you for providing the good work.

I downloaded the pretrained models of SENet and ResNet but when I run inference on CUB and NABirds, the accuracies were different from reported numbers:
Specifically, on CUB dataset, I only got 0.5% compared to 87.x % in your report:

Could you check the provided pretrained files?

Thank you for your help

hi, I want to train CrossX on my dataset, however the results is much worse than resnet's . And I notice that your model have different hyperparameter and structure for different fined-grained dataset, and I want know is there any standard or uniform parameters for other classification dataset?
I would appreciate it if you can reply~

@cswluo I went to the Baidu to get the pretrained models. However, given I am not a chinese user, I was not able to create an account to download. Is it possible to give me access to the nabirds_CrossX-SENet50.model and nabirds_CrossX-SENet50.model please?