Update README.md

README.md

@@ -14,16 +14,6 @@ tags:
 - image
 - depth-estimation
 - DEM
-dataset_info:
-  config_name: default
-  dataset_size: 116710502400
-  splits:
-  - name: train
-    num_bytes: 93757337600
-    num_examples: 65090
-  - name: validation
-    num_bytes: 22953164800
-    num_examples: 15808
 ---
 
 # MCTED - Mars CTX Terrain-Elevation Dataset
@@ -31,17 +21,33 @@ dataset_info:
 ## Overview
 **MCTED** is a machine-learning-ready dataset of optical images of the surface of Mars, paired with their corresponding digital elevation models.
 It was created using an extensive repository of orthoimage-DEM pairs with the NASA Ames Stereo Pipeline using the Mars Reconneissance Orbiter's CTX instrument imagery by
-[Day et al. 2023](https://github.com/GALE-Lab/Mars_DEMs).
+[Day et al. 2023](https://github.com/GALE-Lab/Mars_DEMs). We process the samples from the repository using a developed pipeline aimed at eliminating elevation artifacts, 
+imputing missing data points and sample selection. The dataset is provided in the form of 518x518 patches.
 
 This dataset is fully open-source, with all data and code used for it's generation available publicly.
 
-[**Dataset repository**]() | [**arXiv**]()
+[**Dataset repository**]() | [**arXiv article**]()
 
 ## Dataset contents
-The dataset contains in total **80898** samples, divided into two splits:
+The dataset contains in total **80,898** samples, divided into two splits:
 |Training|Validation|
 |---|---|
-|65090|15808|
+|65,090|15,808|
 
 Each sample consists of 4 different files:
-|
+|Type|Description|
+|---|---|
+|optical.png|The monochromatic optical image patch. Despite being monochromatic, the image still has 3 channels, with all channels being the same|
+|elevation.tiff|The elevation data patch in meters w.r.t. the Martian datum|
+|deviation_mask.png|Binary mask with locations that were identified as elevation artifacts during dataset generation and were replaced with interpolated values|
+|initial_nan_mask.png|Binary mask with locations that contained missing values in the Day et al. data samples and were imputed during processing|
+
+## Data source
+
+The dataset has been generated using a orthoimage-DEM pair repository generated from MROs CTX imagery using the [NASA Ames Stereo Pipeline](https://github.com/NeoGeographyToolkit/StereoPipeline) 
+by [Day et al. 2023](https://faculty.epss.ucla.edu/~mday/index.php/mars-dems/). We pass the samples throgugh an extensive processing and selection pipeline, using approximately **47%** of the available data.
+
+![Sankey diagram of processed samples](./images/sankey_processing.png)
+
+
+## Citation