search

3neutronstar / GuidedMixup

Official PyTorch implementation of "GuidedMixup An Efficient Mixup Strategy Guided by Saliency Maps" (AAAI'23 Oral)
GNU General Public License v3.0
5 stars 4 forks source link

Question about GuidedMixup #1

Closed khawar-islam closed 1 year ago

khawar-islam commented 1 year ago

Dear @3neutronstar ,

I hope this email finds you well. I have a question regarding your GuidedMixup paper. How did you split the Caltech-101 dataset? Are there specific training and testing images?

I would appreciate your prompt response as I need to conduct an experiment urgently.

Best regards,

3neutronstar commented 1 year ago

Dear @khawar-islam ,

Thank you for being so interested in my work.

Yes, there are no specific training and testing images in the official. To this end, I split the Caltech-101 dataset into 0.8:0.2, following the scikit-learn package's train_test_split function. I also use the stratify option using the labels.

khawar-islam commented 1 year ago

Dear @3neutronstar

thank you for your message. The Caltech-101 folder contains 102 folders not 101 folders. What is the reason behind this?

3neutronstar commented 1 year ago

I use Torchvision's Caltech-101 class to train the model. I think it automatically ignores the folder name BACKGROUND_Google in 101_ObjectCategories directory.

khawar-islam commented 1 year ago

@3neutronstar thanks a lot this is a lot for me