Data in paper “Transformer-based Method to Eliminate Cloud Shadow Interference in Automatic Lake Extraction from Sentinel-2 Imagery”

The compressed files are the training datasets, the validation dataset, the lake prediction result and a checkpoint in paper “<strong>Transformer-based Method to Eliminate Cloud Shadow Interference in Automatic Lake Extraction from Sentinel-2 Imagery</strong>”. The “<strong>Training datasets. rar</strong>” includes 5 independent training datasets containing 0%, 1%, 2%, 3% and 4% cloud shadows respectively. Each training dataset includes ground truth folder and remote sensing image folder, namely "gt" and "image", which constitute 5000 sample pairs in total. The “<strong>Validation dataset.rar</strong>” is used to valid the accuracy of the model trained by the five training datasets, also including two folders "gt" and "image" , with a total of 220 sample pairs.The ”<strong>Prediction result.rar</strong>” is the lake prediction result in the Inner River Basin of the Tibetan Plateau using the model trained by the training dataset containing 4% cloud shadows. There are 112 tiles of images covering this region, so a total of 112 folders are included. And the “<strong>BestCheckpoint.ckpt</strong>” file is the model with the highest accuracy in the paper.