Update DATASET_GUIDE.md with xView2

DATASET_GUIDE.md

@@ -15,3 +15,20 @@ This document provides a detailed overview of the datasets used in this reposito
       --augmentation_config "configs/augmentations/segmentation_default.yaml" \
       --use_wandb
   ```
+### xView2
+- The dataset needs to be downloaded manually from the official website. This requires a registration and accepting the terms and conditions. On the download page, we need the datasets under `Datasets from the Challenge`, excluding the holdout set. Extract the datasets in the `./data/xView2/` folder, such that it contains e.g. `./data/xView2/tier3/images/...`.
+- The `tier3` set does not come up labels in the form of images, so we first need to create them from the respective JSON data. We create a `masks` folder on the level of the `images` folder by running:
+
+```
+python datasets/xView2_create_masks.py
+```
+- The basic experimental setup for this dataset is a change detection task. Two images showing the same location are encoded using a foundation model as encoder. A smaller UPerNet model is trained to compute the 5-class segmentation mask from these encodings. Below is a CLI example for running the experiment with the Prithvi pretrained encoder:
+```
+torchrun --nnodes=1 --nproc_per_node=1 run.py  \
+--config configs/run/default.yaml  \
+--encoder_config configs/foundation_models/prithvi.yaml  \
+--dataset_config configs/datasets/xview2.yaml   \
+--segmentor_config configs/segmentors/siamdiffupernet.yaml \
+--augmentation_config configs/augmentations/segmentation_oversampling.yaml  \
+--use_wandb
+```