Update README.md

README.md

@@ -2,16 +2,20 @@
 license: cc-by-4.0
 ---
 
+# BHI_LR_multi
 
-# Scale settings
-[scale]
-# List of available scale algorithms (e.g., down_up,linear,cubic_catrom,cubic_mitchell,cubic_bspline,lanczos,gauss,hamming,hann,box,hermite,lagrange)
-algorithms = down_up,linear,cubic_mitchell,lanczos,gauss,box
-# List of available scale algorithms when applying down_up
-down_up_algorithms = linear,cubic_mitchell,lanczos,gauss,box
-# Whether to choose a random scale algorithm each time (True or False)
-randomize = True
-# Factor to scale your images to (e.g., 0.25, 0.50, 0.75) (0.25 = 25%, 0.50 = 50%)
-size_factor = 0.25
-# Range of values for down_up (e.g., 0.5,2.0) (0.5 = 50%, 2.0 = 200%)
-range = 0.75,1.5
+This is a x4 LR counterpart to the [BHI SISR Dataset](https://huggingface.co/datasets/Phips/BHI/blob/main/README.md).
+
+**SCREENSHOT HERE**
+
+To create this dataset, a 0.25 scaling (or x4 scaling) has been applied, with multiple scaling algorithms in a randomized manner.
+
+The algos used are linear, cubic_mitchell, lanczos, gauss, box and down_up (with these algos).
+The specifically applied degradation for each image can be found in the [applied_degradations.txt](https://huggingface.co/datasets/Phips/BHI_LR_multi/resolve/main/applied_degradations.txt?download=true) file
+
+The trained model learns to handle multiple scalings, and also these counteract each other in the sense of that lanczos will always sharpen, gauss will always soften etc. Using this instead of just bicubic sampling for example will help the model not to be trained on / pick up on one specific scaling algos characteristics.
+
+ <figure>
+  <img src="https://cdn-uploads.huggingface.co/production/uploads/634e9aa407e669188d3912f9/_vs_h4eygsFC27qRdhVha.png" alt="screenshot beginning of applied_degradations.txt">
+  <figcaption>Applied scaling on each image, screenshot beginning of applied_degradations.txt</figcaption>
+</figure>