Skip to content

Synthetic NIR generation

Note

Please keep in mind that if you download all files, they are pretty large in size (~45GB). It is suggested to only download what you need.

Size of each dataset

Dataset Description File Name Size (GB)
nirscene1 x10 oversample nirscene_img_aug_10_oversample_vege_pix2pixHD.tar.gz 0.71
nirscene1 x100 oversample nirscene_img_aug_100_oversample_vege_pix2pixHD.tar.gz 3.66
nirscene1 x200 oversample nirscene_img_aug_200_oversample_vege_pix2pixHD.tar.gz 7.32
nirscene1 x400 oversample nirscene_img_aug_400_oversample_vege_pix2pixHD.tar.gz 14.62
nirscene1 x600 oversample nirscene_img_aug_600_oversample_vege_pix2pixHD.tar.gz 21.93
Dataset Description File Name Size (GB)
SEN12MS Summer sen12ms_SUMMER_pix2pixHD.tar.gz 6.59
SEN12MS All seasons sen12ms_ALL_pix2pixHD.tar.gz 26.7
Dataset Description File Name Size (GB)
capsicum capsicums_pix2pixHD_8_1_1.tar.gz 3.03

Number of data samples

The total number of samples is 355,300 and all datasets are splitted 8:1:1 ratio for train/validation/test. The table below shows more detail information.

Folder structure

Each dataset (e.g., nirscene_img_aug_10_oversample_vege_pix2pixHD.tar.gz) contains 6 folders with suffix _A or _B.

.
└── nirscene_img_aug_pix2pixHD_10_oversample_vege
    ├── test_A
    ├── test_B
    ├── train_A
    ├── train_B
    ├── val_A
    └── val_B
  1. Suffix with _A indicates RGB image
  2. Suffix with _B indicates NIR image
  3. train and val folders include images exploited during training phase and test is used for FID validation.