search

WHU-Sigma / HyperSIGMA

The official repo for the paper "HyperSIGMA: Hyperspectral Intelligence Comprehension Foundation Model"
134 stars 12 forks source link

Test Data Normalization #9

Closed GuochaoCS closed 6 days ago

GuochaoCS commented 1 month ago

During the pretraining phase, all hyperspectral images (HSIs) are normalized by dividing by 4,000. In the context of testing for the super-resolution task, what standardization format should be applied to the Houston dataset? The original Houston dataset contains 50 channels with a spatial resolution of 4172*1202.

leonmakise commented 3 weeks ago

The Houston dataset we offered has been normalized into [0,1], and we use the pre-trained model for initialization when training.

For the testing phase, the images are also pre-processed into [0,1].