search

juhongm999 / hsnet

Official PyTorch Implementation of Hypercorrelation Squeeze for Few-Shot Segmentation, ICCV 2021
231 stars 43 forks source link

VOC dataset mismatching #17

Closed songyinan1 closed 2 years ago

songyinan1 commented 2 years ago

hi, I have noticed that there is a gap between your folds' txt files and the folder "SegmentationClassAug". The total number of masks in "SegmentationClassAug" is 12031, while after removing duplicate files on the folds' TXT statistics of trn and val folder, the total number of image list is only 11473, could you please tell me why don't the numbers match up? Or am I missing some data-processing detail?

juhongm999 commented 2 years ago

We used the image/mask list for each fold from previous work, e.g., PMM repo for fairness in using training/validation data. I'm not sure where the lists are originally from so I suggest you to contact authors of either the original work (OSLSM) or PMM.

songyinan1 commented 2 years ago

We used the image/mask list for each fold from previous work, e.g., PMM repo for fairness in using training/validation data. I'm not sure where the lists are originally from so I suggest you to contact authors of either the original work (OSLSM) or PMM.

Thanks, I will go to check the previous work.