Update README.md

README.md

@@ -21,9 +21,9 @@ size_categories:
 
 
 ## Introduction
-The dataset consists primarily of Sentinel-2 images from 2015 (mainly), 2016 and 2017, and binary segmentation masks for  glaciers, based on an inventory built by glaciology experts (Paul et al. 2020). Secondly, given that glacier ice is not always visible in the images, due to seasonal snow, shadow/cloud cover, and debris cover (which is particularly significant), the dataset also includes additional features (see [additional components](#additional-components)) that can help in the segmentation task. 
+The dataset consists of Sentinel-2 images from 2015 (mainly), 2016 and 2017, and binary segmentation masks for  glaciers, based on an inventory built by glaciology experts (Paul et al. 2020). Secondly, given that glacier ice is not always visible in the images, due to seasonal snow, shadow/cloud cover, and debris cover (which is particularly significant), the dataset also includes additional features (see [additional components](#additional-components)) that can help in the segmentation task. 
 
-The glacier inventory contains 4395 glaciers covering 1805.9 km<sup>2</sup>. We consider only the glaciers larger than 0.1 km<sup>2</sup> (n = 1646 out of 4395). After a manual QC, we additionally removed 53 glaciers for which the images were of poor quality (e.g. too many clouds) and for which alternative dates within the same year could not be found. We end up with 1593 glaciers but covering 93.3% of the glacier inventory area (i.e. 1684.7 km<sup>2</sup>). The dataset was constructed by collecting images for each glacier individually, with a 1.28 km buffer around each glacier (e.g. see [Example for Aletsch Glacier](https://huggingface.co/datasets/dcodrut/dl4gam_alps/resolve/main/preds/g_3073_2015-08-29.png)).
+The glacier inventory contains 4395 glaciers covering 1805.9 km<sup>2</sup>. We consider only the glaciers larger than 0.1 km<sup>2</sup> (n = 1646). After a manual QC, we additionally removed 53 glaciers for which the images were of poor quality (e.g. too many clouds) and for which alternative dates within the same year could not be found. We end up with 1593 glaciers but covering 93.3% (1684.7 km<sup>2</sup>) of the glacier inventory area. The dataset was constructed by collecting images for each glacier individually, with a 1.28 km buffer around each glacier (e.g. see [Example for Aletsch Glacier](https://huggingface.co/datasets/dcodrut/dl4gam_alps/resolve/main/preds/g_3073_2015-08-29.png)).
 
 This repository includes both patchified and glacier-wide raster versions. An example of one patch with all the features:
 <img src="https://cdn-uploads.huggingface.co/production/uploads/655392db8ef4a49e59db8392/g2JwygjGQIZ-IWF0VZDWN.jpeg" alt="Patch example" width="1200px"/>
@@ -65,21 +65,24 @@ Additionally, we provide glacier-wide rasters for 2023, which can be used for gl
 Given each glacier's raster, we sample one or more patches, of size 256x256 (i.e. 2.56 km x 2.56 km) as follows:
 - version `small` (2251 patches): 
   - we sample one patch with the center on the glacier's centroid
-  - we build a grid of patch centers with a step size of 128 pixels and keep all the patches which have the center  on glacier
+  - we build a grid of patch centers with a step size of 128 pixels and keep all the patches which have the center on the (current) glacier
 - version `large` (11440 patches, used in [our paper](https://doi.org/10.22541/essoar.173557607.70204641/v1)):
   - additional to the centroid, we sample four patches from the "edges"
-  - we use again a grid sampling but with a 64 pixels step 
+  - we use again the grid sampling but with a 64 pixels step 
 
 We show below an example of the two versions of sampled patches (together with their centers) for one glacier ([Glacier de la Pilatte](https://commons.wikimedia.org/wiki/Category:Glacier_de_la_Pilatte)):
 <img src="https://cdn-uploads.huggingface.co/production/uploads/655392db8ef4a49e59db8392/9sQ0_5xhVo2kMdn2TDnvh.png" alt="Patch sampling" width="800px"/>
 
+
 ## Geographic cross-validation
 
-Given that we are interested in glacier area change estimation and we want estimates for all the glaciers, we use a five-fold cross-validation scheme such that we avoid using the training results. Then, a further split is performed in each iteration to extract a small validation fold. The scheme is depicted below, followed by the map of the first iteration:
+Given that we are interested in glacier area change estimation and we want estimates for all the glaciers, we use a five-fold cross-validation scheme such that we avoid using the training results. Then, a further split is performed in each iteration to extract a small validation fold. This acts as a spatial gap between training and testing folds (otherwise it's difficult to have a perfect separation given the irregularity of the glacier shapes). 
+The IDs of the glaciers from each fold and cross-validation iteration are provided in [map_all_splits_all_folds.csv](https://huggingface.co/datasets/dcodrut/dl4gam_alps/blob/main/data/map_all_splits_all_folds.csv).
+The scheme is depicted below, followed by the map of the first iteration:
 <img src="https://cdn-uploads.huggingface.co/production/uploads/655392db8ef4a49e59db8392/SXx9nksbdo5O5rAplL4MU.png" alt="Geographic cross-validation scheme" width="800px"/>
 <img src="https://cdn-uploads.huggingface.co/production/uploads/655392db8ef4a49e59db8392/V7WWMRB2Tb81LLFXdFbN3.png" alt="Geographic cross-validation map (first iteration)" width="1200px"/>
 
-The IDs of the glaciers from each fold and cross-validation iteration are provided in [map_all_splits_all_folds.csv](https://huggingface.co/datasets/dcodrut/dl4gam_alps/blob/main/data/map_all_splits_all_folds.csv).
+
 
 ## How to cite