Add usage instructions and examples of .

README.md

@@ -21,7 +21,7 @@ task_ids: []
 viewer: false
 ---
 
-_This custom Dataset Teaser is just so much better, no wonder it was disabled!_
+_The Dataset Teaser is now enabled instead! Isn't this better?_
 
 ![preview of all texture sets](https://huggingface.co/datasets/texturedesign/td01_natural-ground-textures/resolve/main/docs/teaser.webp)
 
@@ -38,8 +38,33 @@ Overall information about this dataset:
 * **Technique** — hand-held
 * **Orientation** — portrait or landscape
 * **Author**: Alex J. Champandard
+* **Configurations**: 4K, 2K (default), 1K
 
-Photos are arranged into multiple sets...
+To load the medium- and high-resolution images of the dataset, you'll need to install `jxlpy` from [PyPI](https://pypi.org/project/jxlpy/) with `pip install jxlpy`:
+
+```python
+# Recommended use, JXL at high-quality.
+from jxlpy import JXLImagePlugin
+from datasets import load_dataset
+d = load_dataset('texturedesign/td01_natural-ground-textures', 'JXL@4K')
+
+print(len(d['train']), len(d['test']))
+```
+
+The lowest-resolution images are available as PNG with a regular installation of `pillow`:
+
+```python
+# Alternative use, PNG at low-quality.
+from datasets import load_dataset
+d = load_dataset('texturedesign/td01_natural-ground-textures', 'PNG@1K')
+
+# EXAMPLE: Discard all other sets except Set #1.
+dataset = dataset.filter(lambda s: s['set'] == 1)
+# EXAMPLE: Only keep images with index 0 and 2.
+dataset = dataset.select([0, 2])
+```
+
+Use built-in dataset `filter()` and `select()` to narrow down the loaded dataset for training, or to ease with development.
 
 
 ## Set #1: Rock and Gravel