search

YapengMeng / SpecGAN

The official PyTorch implementation of SpecGAN, Large-factor Super-resolution of Remote Sensing Images with Spectra-guided Generative Adversarial Networks (Accept by TGRS).
19 stars 0 forks source link

issues about datasets #1

Open FlamePanda opened 1 year ago

FlamePanda commented 1 year ago

Hi! your work is very nice! I want to reproduce your work, but I don't have your datasets setting, can you list the detailed datasets settings? Thank you very much if you can.

YapengMeng commented 1 year ago

Our dataset preprocessing is a bit complicated, including 1) reshaping images (HSI:601*596->376*376, RGBHR:11920*12020->12032*12032, where 12032=376*32, for ×32 SR), 2)registration by SIFT method, using RGB band in HSI (band 23, 12, 5 in HSI) to matching RGBHR, 3)cut each large image into 4 middle size images, and split for train, valid and test dataset (6:1:1), 4)delete black part (Some parts of the original data set are missing, in RGBHR, where looks black), 5)cut middle size image into small images, for the training set, random cut 78000 pieces, for the valid and test dataset, cut the middle size image in sequence.

You can get our whole dataset from here: https://pan.baidu.com/s/17Gqvu7kzvxSZETqAA1pTww?pwd=mj5y

paolodep36 commented 1 week ago

Dear @YapengMeng could you upload the dataset on drive? Unfortunately i can't access to baidu Thanks in advance