Update DATASET_GUIDE.md

DATASET_GUIDE.md

@@ -39,5 +39,18 @@ This document provides a detailed overview of the datasets used in this reposito
 - The basic experimental setup for this dataset is a multi-temporal multi-modal semantic segmentation task. For models that don't support multi-temporal data, each time frame is processed separately for feature extraction and then mapped into a single representation. This setup requires the configuration file `configs/segmentors/upernet_mt.yaml`. Additionally, in the dataset configuration, specify the number of time frames, for example, `multi_temporal: 6`, where the latest six images are selected for both optical and SAR data. Below is a CLI example for running the experiment using the CROMA pretrained encoder, which jointly processes optical and SAR information:
 
   ```
-  
+###  AI4SmallFarms
+- The code supports automatic downloading of the dataset into `./data` folder.
+- The original dataset contains vector files as well as Google Maps (GM) files, which are skipped during the experiment. Only the .tif Sentinel-2 images and delineation labels are kept after downloading.
+- The dataset is uni-temporal, and the labels contain only two classes (farm boundary or background). For training using the Prithvi encoder, the following command should be used:
+  ```
+  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/ai4smallfarms.yaml" \
+  --segmentor_config "configs/segmentors/upernet_binary.yaml" \
+  --augmentation_config "configs/augmentations/ai4smallfarms.yaml" \
+  --use_wandb
+  ```
+
   ```