Add SenBench-LC100Cls-S3 dataset and SenBench-LC100Seg-S3 dataset from SentinelBench

[wangyi111]

This is a draft PR adding SenBench-LC100Cls-S3 and SenBench-LC100Seg-S3 datasets from SentinelBench. The two datasets share images and labels.

SenBench-LC100Cls-S3

The SenBench-LC100Cls-S3 dataset is a level-2 dataset from the SentinelBench benchmark.
It contains Sentinel-3 OLCI images and multi-label land cover labels for the land cover classification task.
It supports both static (1 image / location) and time series (1-4 images / location) mode, the former is used in the original benchmark.

Dataset features:

* task: multi-label classification
* # samples: 5181/1727/1727 (train/val/test, static mode)
* image resolution: 96x96 (GSD 300m)
* # classes: 23


Dataset format:

* images: 21-band Sentinel-3 OLCI images (GeoTIFF)
* labels: multi-label land cover labels (CSV)

SenBench-LC100Seg-S3

The SenBench-LC100Seg-S3 dataset is a level-2 dataset from the SentinelBench benchmark.
It contains Sentinel-3 OLCI images and land cover maps for the land cover segmentation task.
It supports both static (1 image / location) and time series (1-4 images / location) mode, the former is used in the original benchmark.

Dataset features:

* task: semantic segmentation
* # samples: 5181/1727/1727 (train/val/test, static mode)
* image resolution: 96x96 (GSD 300m)
* label resolution: 282x282 (GSD 100m)
* # classes: 23


Dataset format:

* images: 21-band Sentinel-3 OLCI images (GeoTIFF)
* labels: land cover maps (GeoTIFF)

[wangyi111]

this is based on the previous PR, maybe it's better to keep them separate and all base on the main..